Add files using upload-large-folder tool

.gitattributes

@@ -1,35 +1,3 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
 *.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+target/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+assistant/tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,106 @@
+---
+license: apache-2.0
+license_link: https://ai.google.dev/gemma/docs/gemma_4_license
+base_model:
+- google/gemma-4-31B-it
+- google/gemma-4-31B-it-assistant
+library_name: mlx
+tags:
+- mlx
+- gemma4
+- mtplx
+- speculative-decoding
+- apple-silicon
+- text-generation
+pipeline_tag: text-generation
+---
+
+# Gemma4 MTPLX Optimized Speed
+
+This is an **MTPLX pair bundle** for Gemma 4 31B speculative decoding on Apple Silicon.
+
+It is not a single vanilla Transformers model directory. The repository contains two MLX-format artifacts:
+
+- `target/` - Gemma 4 31B IT target, MLX Q4 affine group-size 64
+- `assistant/` - official Gemma 4 31B assistant drafter, MLX Q6 affine group-size 64
+
+Use this pair when absolute throughput is the priority.
+
+## Source
+
+- Target source: `google/gemma-4-31B-it`
+- Target revision: `145dc2508c480a64b47242f160d286cff94a2343`
+- Assistant source: `google/gemma-4-31B-it-assistant`
+- Assistant revision: `cffbbd2cea41ea56a0fa5b0487e0d445121fd204`
+
+Both artifacts were converted locally to MLX format.
+
+## Quantization
+
+Target:
+
+```text
+bits: 4
+group_size: 64
+mode: affine
+```
+
+Assistant:
+
+```text
+bits: 6
+group_size: 64
+mode: affine
+```
+
+## MTPLX Usage
+
+After downloading this repository, point MTPLX at the two subdirectories:
+
+```bash
+mtplx bench gemma-mtp \
+  --target-model ./target \
+  --assistant-model ./assistant \
+  --prompt-suite mtplx/benchmarks/prompts/flappy.jsonl \
+  --max-tokens 1000 \
+  --draft-block-sizes 6 \
+  --allow-unverified-gemma
+```
+
+The Gemma 4 assistant is a separate drafter model. MTPLX uses exact speculative sampling with target verification and residual correction.
+
+## Local Benchmark
+
+Prompt: single-file HTML5 Canvas Flappy Bird game, capped at 1000 generated tokens.
+
+Sampler:
+
+```text
+temperature: 1.0
+top_p: 0.95
+top_k: 64
+seed: 0
+```
+
+Best observed block size:
+
+```text
+block_size: 6
+acceptance: 830 / 846 = 98.11%
+```
+
+Observed MTPLX throughput samples:
+
+```text
+43.56 tok/s
+44.46 tok/s
+44.07 tok/s
+```
+
+The bundled benchmark JSON files are in `benchmarks/`.
+
+## Notes
+
+This release is optimized for MTPLX speed experiments. For a higher-precision target, use `Youssofal/Gemma4-MTPLX-Optimized-Quality`.
+
+Gemma 4 is released by Google under the Gemma 4 license terms linked above.

assistant/README.md

@@ -0,0 +1,7 @@
+---
+language: en
+library_name: mlx
+pipeline_tag: text-generation
+tags:
+- mlx
+---

assistant/config.json

@@ -0,0 +1,103 @@
+{
+    "architectures": [
+        "Gemma4AssistantForCausalLM"
+    ],
+    "audio_token_id": 258881,
+    "backbone_hidden_size": 5376,
+    "boa_token_id": 256000,
+    "boi_token_id": 255999,
+    "centroid_intermediate_top_k": 32,
+    "dtype": "bfloat16",
+    "eoa_token_id": 258883,
+    "eoi_token_id": 258882,
+    "eos_token_id": [
+        1,
+        106,
+        50
+    ],
+    "image_token_id": 258880,
+    "model_type": "gemma4_assistant",
+    "num_centroids": 2048,
+    "quantization": {
+        "group_size": 64,
+        "bits": 6,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 6,
+        "mode": "affine"
+    },
+    "text_config": {
+        "_name_or_path": "",
+        "architectures": null,
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attention_k_eq_v": true,
+        "bos_token_id": 2,
+        "chunk_size_feed_forward": 0,
+        "dtype": "bfloat16",
+        "enable_moe_block": false,
+        "eos_token_id": 1,
+        "final_logit_softcapping": null,
+        "global_head_dim": 512,
+        "head_dim": 256,
+        "hidden_activation": "gelu_pytorch_tanh",
+        "hidden_size": 1024,
+        "hidden_size_per_layer_input": 0,
+        "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+        },
+        "initializer_range": 0.02,
+        "intermediate_size": 8192,
+        "is_encoder_decoder": false,
+        "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+        },
+        "layer_types": [
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention"
+        ],
+        "max_position_embeddings": 262144,
+        "model_type": "gemma4_text",
+        "moe_intermediate_size": null,
+        "num_attention_heads": 32,
+        "num_experts": null,
+        "num_global_key_value_heads": 4,
+        "num_hidden_layers": 4,
+        "num_key_value_heads": 16,
+        "num_kv_shared_layers": 4,
+        "output_attentions": false,
+        "output_hidden_states": false,
+        "pad_token_id": 0,
+        "problem_type": null,
+        "return_dict": true,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "full_attention": {
+                "partial_rotary_factor": 0.25,
+                "rope_theta": 1000000.0,
+                "rope_type": "proportional"
+            },
+            "sliding_attention": {
+                "rope_theta": 10000.0,
+                "rope_type": "default"
+            }
+        },
+        "sliding_window": 1024,
+        "tie_word_embeddings": true,
+        "top_k_experts": null,
+        "use_bidirectional_attention": null,
+        "use_cache": true,
+        "use_double_wide_mlp": false,
+        "vocab_size": 262144,
+        "vocab_size_per_layer_input": 0
+    },
+    "tie_word_embeddings": true,
+    "transformers_version": "5.7.0.dev0",
+    "use_ordered_embeddings": false
+}

assistant/generation_config.json

@@ -0,0 +1,17 @@
+{
+  "bos_token_id": 2,
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106,
+    50
+  ],
+  "is_assistant": true,
+  "num_assistant_tokens": 6,
+  "num_assistant_tokens_schedule": "constant",
+  "pad_token_id": 0,
+  "temperature": 1.0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.7.0.dev0"
+}

assistant/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:125bbd650ad1b7223b521ff2db20c8f54bc9ce6d4c16a02342fa48c4b9b6fa81
+size 381516281

assistant/model.safetensors.index.json

@@ -0,0 +1,102 @@
+{
+    "metadata": {
+        "total_size": 381506056,
+        "total_parameters": 469518592
+    },
+    "weight_map": {
+        "model.embed_tokens.biases": "model.safetensors",
+        "model.embed_tokens.scales": "model.safetensors",
+        "model.embed_tokens.weight": "model.safetensors",
+        "model.layers.0.input_layernorm.weight": "model.safetensors",
+        "model.layers.0.layer_scalar": "model.safetensors",
+        "model.layers.0.mlp.down_proj.biases": "model.safetensors",
+        "model.layers.0.mlp.down_proj.scales": "model.safetensors",
+        "model.layers.0.mlp.down_proj.weight": "model.safetensors",
+        "model.layers.0.mlp.gate_proj.biases": "model.safetensors",
+        "model.layers.0.mlp.gate_proj.scales": "model.safetensors",
+        "model.layers.0.mlp.gate_proj.weight": "model.safetensors",
+        "model.layers.0.mlp.up_proj.biases": "model.safetensors",
+        "model.layers.0.mlp.up_proj.scales": "model.safetensors",
+        "model.layers.0.mlp.up_proj.weight": "model.safetensors",
+        "model.layers.0.post_attention_layernorm.weight": "model.safetensors",
+        "model.layers.0.post_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.0.pre_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.0.self_attn.o_proj.biases": "model.safetensors",
+        "model.layers.0.self_attn.o_proj.scales": "model.safetensors",
+        "model.layers.0.self_attn.o_proj.weight": "model.safetensors",
+        "model.layers.0.self_attn.q_norm.weight": "model.safetensors",
+        "model.layers.0.self_attn.q_proj.biases": "model.safetensors",
+        "model.layers.0.self_attn.q_proj.scales": "model.safetensors",
+        "model.layers.0.self_attn.q_proj.weight": "model.safetensors",
+        "model.layers.1.input_layernorm.weight": "model.safetensors",
+        "model.layers.1.layer_scalar": "model.safetensors",
+        "model.layers.1.mlp.down_proj.biases": "model.safetensors",
+        "model.layers.1.mlp.down_proj.scales": "model.safetensors",
+        "model.layers.1.mlp.down_proj.weight": "model.safetensors",
+        "model.layers.1.mlp.gate_proj.biases": "model.safetensors",
+        "model.layers.1.mlp.gate_proj.scales": "model.safetensors",
+        "model.layers.1.mlp.gate_proj.weight": "model.safetensors",
+        "model.layers.1.mlp.up_proj.biases": "model.safetensors",
+        "model.layers.1.mlp.up_proj.scales": "model.safetensors",
+        "model.layers.1.mlp.up_proj.weight": "model.safetensors",
+        "model.layers.1.post_attention_layernorm.weight": "model.safetensors",
+        "model.layers.1.post_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.1.pre_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.1.self_attn.o_proj.biases": "model.safetensors",
+        "model.layers.1.self_attn.o_proj.scales": "model.safetensors",
+        "model.layers.1.self_attn.o_proj.weight": "model.safetensors",
+        "model.layers.1.self_attn.q_norm.weight": "model.safetensors",
+        "model.layers.1.self_attn.q_proj.biases": "model.safetensors",
+        "model.layers.1.self_attn.q_proj.scales": "model.safetensors",
+        "model.layers.1.self_attn.q_proj.weight": "model.safetensors",
+        "model.layers.2.input_layernorm.weight": "model.safetensors",
+        "model.layers.2.layer_scalar": "model.safetensors",
+        "model.layers.2.mlp.down_proj.biases": "model.safetensors",
+        "model.layers.2.mlp.down_proj.scales": "model.safetensors",
+        "model.layers.2.mlp.down_proj.weight": "model.safetensors",
+        "model.layers.2.mlp.gate_proj.biases": "model.safetensors",
+        "model.layers.2.mlp.gate_proj.scales": "model.safetensors",
+        "model.layers.2.mlp.gate_proj.weight": "model.safetensors",
+        "model.layers.2.mlp.up_proj.biases": "model.safetensors",
+        "model.layers.2.mlp.up_proj.scales": "model.safetensors",
+        "model.layers.2.mlp.up_proj.weight": "model.safetensors",
+        "model.layers.2.post_attention_layernorm.weight": "model.safetensors",
+        "model.layers.2.post_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.2.pre_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.2.self_attn.o_proj.biases": "model.safetensors",
+        "model.layers.2.self_attn.o_proj.scales": "model.safetensors",
+        "model.layers.2.self_attn.o_proj.weight": "model.safetensors",
+        "model.layers.2.self_attn.q_norm.weight": "model.safetensors",
+        "model.layers.2.self_attn.q_proj.biases": "model.safetensors",
+        "model.layers.2.self_attn.q_proj.scales": "model.safetensors",
+        "model.layers.2.self_attn.q_proj.weight": "model.safetensors",
+        "model.layers.3.input_layernorm.weight": "model.safetensors",
+        "model.layers.3.layer_scalar": "model.safetensors",
+        "model.layers.3.mlp.down_proj.biases": "model.safetensors",
+        "model.layers.3.mlp.down_proj.scales": "model.safetensors",
+        "model.layers.3.mlp.down_proj.weight": "model.safetensors",
+        "model.layers.3.mlp.gate_proj.biases": "model.safetensors",
+        "model.layers.3.mlp.gate_proj.scales": "model.safetensors",
+        "model.layers.3.mlp.gate_proj.weight": "model.safetensors",
+        "model.layers.3.mlp.up_proj.biases": "model.safetensors",
+        "model.layers.3.mlp.up_proj.scales": "model.safetensors",
+        "model.layers.3.mlp.up_proj.weight": "model.safetensors",
+        "model.layers.3.post_attention_layernorm.weight": "model.safetensors",
+        "model.layers.3.post_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.3.pre_feedforward_layernorm.weight": "model.safetensors",
+        "model.layers.3.self_attn.o_proj.biases": "model.safetensors",
+        "model.layers.3.self_attn.o_proj.scales": "model.safetensors",
+        "model.layers.3.self_attn.o_proj.weight": "model.safetensors",
+        "model.layers.3.self_attn.q_norm.weight": "model.safetensors",
+        "model.layers.3.self_attn.q_proj.biases": "model.safetensors",
+        "model.layers.3.self_attn.q_proj.scales": "model.safetensors",
+        "model.layers.3.self_attn.q_proj.weight": "model.safetensors",
+        "model.norm.weight": "model.safetensors",
+        "post_projection.biases": "model.safetensors",
+        "post_projection.scales": "model.safetensors",
+        "post_projection.weight": "model.safetensors",
+        "pre_projection.biases": "model.safetensors",
+        "pre_projection.scales": "model.safetensors",
+        "pre_projection.weight": "model.safetensors"
+    }
+}

assistant/mtplx_artifact.json

@@ -0,0 +1,12 @@
+{
+  "note": "Converted locally with MTPLX Gemma4 assistant classes because stock mlx_lm.convert does not support model_type=gemma4_assistant.",
+  "precision_policy": "Q6 affine G64 for all quantizable assistant modules, including tied embedding/LM-head path and projections.",
+  "quantization": {
+    "bits": 6,
+    "group_size": 64,
+    "mode": "affine"
+  },
+  "source_path": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-bf16-mlx",
+  "source_repo": "google/gemma-4-31B-it-assistant",
+  "source_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204"
+}

assistant/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75a6583c1a418e2bbd79c60d95d28e0f5bf549ad3f2990b5bdb5238c6c2bf70c
+size 32169440

assistant/tokenizer_config.json

@@ -0,0 +1,52 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [],
+  "image_token": "<|image|>",
+  "is_local": true,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>"
+}

benchmarks/flappy1000-targetq4-assistantq6-confirm-block6.json

@@ -0,0 +1,1183 @@
+{
+  "arch_id": "gemma4-assistant-mtp",
+  "artifacts": {
+    "assistant_dtype": null,
+    "assistant_format": "q6-g64-affine",
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "assistant_quantization": {
+      "bits": 6,
+      "group_size": 64,
+      "mode": "affine"
+    },
+    "disk_ok": true,
+    "min_free_gib": 220.0,
+    "observed_free_gib": 704.1293067932129,
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+    "target_quantization": {
+      "bits": 4,
+      "format": "mlx-flat4-g64",
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "backend": "gemma4_assistant",
+  "benchmark": {
+    "draft_block_sizes": [
+      6
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "max_mode": true,
+    "max_tokens": 1000,
+    "profile": "sustained",
+    "prompt_suite": "mtplx/benchmarks/prompts/flappy.jsonl",
+    "reasoning": "off",
+    "sampler_source": {
+      "do_sample": true,
+      "local_reference": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx/generation_config.json",
+      "name": "official Gemma 4 generation_config.json",
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    },
+    "seed": 0,
+    "temperature": 1.0,
+    "top_k": 64,
+    "top_p": 0.95
+  },
+  "blockers": [
+    "best speedup 1.852x is below 2.000x",
+    "median confirmation speedup 1.852x is below 2.000x"
+  ],
+  "can_run_now": true,
+  "gates": {
+    "generated_tokens": 1000,
+    "longer_lengths_blocked_until_160_passes": true,
+    "median_of_3_min_speedup_vs_ar": 2.0,
+    "min_speedup_vs_ar": 2.0,
+    "mtp_peak_memory_lte_ar_multiplier": 1.18,
+    "mtp_peak_memory_lte_ar_plus_gib": 6
+  },
+  "official_sources": {
+    "assistant": "google/gemma-4-31B-it-assistant",
+    "assistant_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204",
+    "target": "google/gemma-4-31B-it",
+    "target_revision": "145dc2508c480a64b47242f160d286cff94a2343"
+  },
+  "pair": {
+    "assistant_exists": true,
+    "assistant_inspection": {
+      "architecture": "Gemma4AssistantForCausalLM",
+      "architecture_recognized": true,
+      "backbone_hidden_size": 5376,
+      "compatibility": {
+        "arch_id": "gemma4-assistant-mtp",
+        "can_run": false,
+        "exit_code": 3,
+        "message": "Official-style Gemma 4 31B assistant artifact recognized. This is an assistant-backed MTP pair, not a standalone target; MTPLX scaffold is present but QA and the 160-token speed/memory gate are still pending.",
+        "mtp_supported": "recognized",
+        "recognized": true,
+        "recommended_backend": "gemma4_assistant",
+        "recommended_profile": "performance-cold",
+        "runtime_compatibility": "assistant-pair-qa-pending",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "architecture-scaffolded-qa-pending",
+        "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+        "supported": false,
+        "tier": "architecture-compatible-but-unverified",
+        "unsafe_force_required": false,
+        "unverified_model": true
+      },
+      "config_exists": true,
+      "hidden_size": 1024,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+      "model_files": [
+        "model.safetensors"
+      ],
+      "model_type": "gemma4_assistant",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": "gemma4-assistant-mtp",
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "recognized",
+      "num_hidden_layers": 4,
+      "num_kv_shared_layers": 4,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 6,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": "gemma4_assistant",
+      "recommended_profile": "performance-cold",
+      "runtime_compatibility": "assistant-pair-qa-pending",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "architecture-scaffolded-qa-pending",
+      "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+      "unverified_model": true,
+      "use_ordered_embeddings": false,
+      "vocab_size": 262144
+    },
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "pair_error": null,
+    "pair_valid": true,
+    "target_exists": true,
+    "target_inspection": {
+      "architecture": "Gemma4ForConditionalGeneration",
+      "architecture_recognized": false,
+      "backbone_hidden_size": null,
+      "compatibility": {
+        "arch_id": null,
+        "can_run": false,
+        "exit_code": 2,
+        "message": "Model has no MTP head. MTPLX requires an MTP-equipped model.",
+        "mtp_supported": "no",
+        "recognized": false,
+        "recommended_backend": null,
+        "recommended_profile": null,
+        "runtime_compatibility": "unsupported",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "unsupported",
+        "support_notes": null,
+        "supported": false,
+        "tier": "no-MTP",
+        "unsafe_force_required": false,
+        "unverified_model": false
+      },
+      "config_exists": true,
+      "hidden_size": 5376,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+      "model_files": [
+        "model-00001-of-00004.safetensors",
+        "model-00002-of-00004.safetensors",
+        "model-00003-of-00004.safetensors",
+        "model-00004-of-00004.safetensors"
+      ],
+      "model_type": "gemma4_text",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": null,
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "no",
+      "num_hidden_layers": 60,
+      "num_kv_shared_layers": 0,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 4,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": null,
+      "recommended_profile": null,
+      "runtime_compatibility": "unsupported",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "unsupported",
+      "support_notes": null,
+      "unverified_model": false,
+      "use_ordered_embeddings": null,
+      "vocab_size": 262144
+    },
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx"
+  },
+  "planned_commands": {
+    "assistant_bf16_snapshot": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it-assistant', revision='cffbbd2cea41ea56a0fa5b0487e0d445121fd204', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx')\"",
+    "gate": "mtplx bench gemma-mtp --target-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --assistant-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx --profile sustained --max --prompt-suite mtplx/benchmarks/prompts/flappy.jsonl --max-tokens 1000 --temperature 1.0 --top-p 0.95 --top-k 64 --seed 0 --reasoning off --draft-block-sizes 6  --json --output outputs/gemma4/flappy1000-q6assistant-pure-confirm-block6.json",
+    "target_flat4_g64": "uv run python -m mlx_lm.convert --hf-path /Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48 --mlx-path /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --quantize --q-bits 4 --q-group-size 64 --q-mode affine",
+    "target_revision_download": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it', revision='145dc2508c480a64b47242f160d286cff94a2343', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48')\""
+  },
+  "qa_pending": true,
+  "results": {
+    "ar": {
+      "active_memory_gib": 16.96542397607118,
+      "cache_memory_gib": 1.9759961115196347,
+      "decode_s": 41.648729124994134,
+      "generated_tokens": 1000,
+      "mode": "ar",
+      "peak_memory_gib": 17.040895924903452,
+      "prefill_s": 0.7671576249995269,
+      "tok_s": 24.010336473865717,
+      "token_preview": [
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761
+      ]
+    },
+    "ar_confirmation": [
+      {
+        "active_memory_gib": 16.96542397607118,
+        "cache_memory_gib": 1.9759961115196347,
+        "decode_s": 41.648729124994134,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 17.040895924903452,
+        "prefill_s": 0.7671576249995269,
+        "tok_s": 24.010336473865717,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      }
+    ],
+    "best_block_confirmation": [
+      {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.410899550653994,
+        "block_size": 6,
+        "cache_memory_gib": 67.3336522737518,
+        "decode_s": 22.49344370800827,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566339950077236,
+        "prefill_s": 0.2826114170020446,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.8515941651996162,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.015003502820036374,
+          "draft": 1.5297440800641198,
+          "rollback": 0.00013703109289053828,
+          "target_distribution": 20.73823187011294,
+          "target_hidden": 0.0,
+          "verify": 0.1980874161090469
+        },
+        "tok_s": 44.457398919489286,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    ],
+    "best_block_size": 6,
+    "best_speedup": 1.8515941651996162,
+    "blockers": [
+      "best speedup 1.852x is below 2.000x",
+      "median confirmation speedup 1.852x is below 2.000x"
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "fan_restore": {
+      "after": {
+        "actual_max_rpm": 5766,
+        "actual_min_rpm": 5419,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 5419,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5419,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 4818
+            },
+            "rpm": 5419,
+            "target_rpm": 4818
+          },
+          {
+            "actual_rpm": 5766,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5766,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5203
+            },
+            "rpm": 5766,
+            "target_rpm": 5203
+          }
+        ],
+        "max_rpm": 5766,
+        "min_rpm": 5419,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5419,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4818\n    },\n    {\n      \"actual_rpm\" : 5766,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5203\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.1,\n    \"TCDX\" : 75.2,\n    \"TCHP\" : 61.6,\n    \"TCMb\" : 89.2,\n    \"TG0B\" : 32.1,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.1,\n    \"TH0x\" : 37.9,\n    \"TMVR\" : 67,\n    \"TPDX\" : 67.3,\n    \"TRDX\" : 80.6,\n    \"TS0P\" : 73.2,\n    \"Tg0j\" : 78.6,\n    \"Tm08\" : 73.9,\n    \"Tp04\" : 75.3,\n    \"Tp08\" : 74.8,\n    \"Tp0C\" : 76.3,\n    \"Tp0G\" : 76.4,\n    \"Tp0X\" : 75.9\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5419,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4818\n    },\n    {\n      \"actual_rpm\" : 5766,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5203\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.1,\n    \"TCDX\" : 75.2,\n    \"TCHP\" : 61.6,\n    \"TCMb\" : 89.2,\n    \"TG0B\" : 32.1,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.1,\n    \"TH0x\" : 37.9,\n    \"TMVR\" : 67,\n    \"TPDX\" : 67.3,\n    \"TRDX\" : 80.6,\n    \"TS0P\" : 73.2,\n    \"Tg0j\" : 78.6,\n    \"Tm08\" : 73.9,\n    \"Tp04\" : 75.3,\n    \"Tp08\" : 74.8,\n    \"Tp0C\" : 76.3,\n    \"Tp0G\" : 76.4,\n    \"Tp0X\" : 75.9\n  }\n}"
+          }
+        },
+        "target_max_rpm": 5203,
+        "target_min_rpm": 4818
+      },
+      "message": "fan profile restored",
+      "ok": true,
+      "profile": "silent",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "auto"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "No matching processes were found",
+            "stdout": "Fans reset to Apple defaults"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "auto"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "silent"
+      }
+    },
+    "fanmax": {
+      "after": {
+        "actual_max_rpm": 7406,
+        "actual_min_rpm": 7372,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 7372,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7372,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7372,
+            "target_rpm": 7826
+          },
+          {
+            "actual_rpm": 7406,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7406,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7406,
+            "target_rpm": 7826
+          }
+        ],
+        "max_rpm": 7406,
+        "min_rpm": 7372,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7372,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7406,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 31.9,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 56.9,\n    \"TCMb\" : 72.3,\n    \"TG0B\" : 31.9,\n    \"TG0H\" : 31,\n    \"TG0V\" : 31.9,\n    \"TH0x\" : 37.2,\n    \"TMVR\" : 56.2,\n    \"TPDX\" : 61.2,\n    \"TRDX\" : 63.1,\n    \"TS0P\" : 63.3,\n    \"Tg0j\" : 62.4,\n    \"Tm08\" : 62.5,\n    \"Tp04\" : 63.4,\n    \"Tp08\" : 62.8,\n    \"Tp0C\" : 63.7,\n    \"Tp0G\" : 64.2,\n    \"Tp0X\" : 63.2\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7372,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7406,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 31.9,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 56.9,\n    \"TCMb\" : 72.3,\n    \"TG0B\" : 31.9,\n    \"TG0H\" : 31,\n    \"TG0V\" : 31.9,\n    \"TH0x\" : 37.2,\n    \"TMVR\" : 56.2,\n    \"TPDX\" : 61.2,\n    \"TRDX\" : 63.1,\n    \"TS0P\" : 63.3,\n    \"Tg0j\" : 62.4,\n    \"Tm08\" : 62.5,\n    \"Tp04\" : 63.4,\n    \"Tp08\" : 62.8,\n    \"Tp0C\" : 63.7,\n    \"Tp0G\" : 64.2,\n    \"Tp0X\" : 63.2\n  }\n}"
+          }
+        },
+        "target_max_rpm": 7826,
+        "target_min_rpm": 7826
+      },
+      "baseline": {
+        "actual_max_rpm": 5219,
+        "actual_min_rpm": 4823,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 4823,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 4823,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 4826
+            },
+            "rpm": 4823,
+            "target_rpm": 4826
+          },
+          {
+            "actual_rpm": 5219,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5219,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5212
+            },
+            "rpm": 5219,
+            "target_rpm": 5212
+          }
+        ],
+        "max_rpm": 5219,
+        "min_rpm": 4823,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 4823,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4826\n    },\n    {\n      \"actual_rpm\" : 5219,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5212\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 31.9,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 57.2,\n    \"TCMb\" : 73.1,\n    \"TG0B\" : 31.9,\n    \"TG0H\" : 31,\n    \"TG0V\" : 31.9,\n    \"TH0x\" : 37.3,\n    \"TMVR\" : 56.7,\n    \"TPDX\" : 61.7,\n    \"TRDX\" : 63.6,\n    \"TS0P\" : 63.9,\n    \"Tg0j\" : 62.9,\n    \"Tm08\" : 62.9,\n    \"Tp04\" : 63.7,\n    \"Tp08\" : 63,\n    \"Tp0C\" : 63.9,\n    \"Tp0G\" : 64.5,\n    \"Tp0X\" : 63.6\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 4823,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4826\n    },\n    {\n      \"actual_rpm\" : 5219,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5212\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 31.9,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 57.2,\n    \"TCMb\" : 73.1,\n    \"TG0B\" : 31.9,\n    \"TG0H\" : 31,\n    \"TG0V\" : 31.9,\n    \"TH0x\" : 37.3,\n    \"TMVR\" : 56.7,\n    \"TPDX\" : 61.7,\n    \"TRDX\" : 63.6,\n    \"TS0P\" : 63.9,\n    \"Tg0j\" : 62.9,\n    \"Tm08\" : 62.9,\n    \"Tp04\" : 63.7,\n    \"Tp08\" : 63,\n    \"Tp0C\" : 63.9,\n    \"Tp0G\" : 64.5,\n    \"Tp0X\" : 63.6\n  }\n}"
+          }
+        },
+        "target_max_rpm": 5212,
+        "target_min_rpm": 4826
+      },
+      "message": "fans ramped to max (actual 7372-7406 RPM; target 7826 RPM)",
+      "ok": true,
+      "profile": "max",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "max"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "Fan 0: 4823 RPM \u2192 max (7826 RPM)\nFan 1: 5219 RPM \u2192 max (7826 RPM)"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "max"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "max"
+      }
+    },
+    "max_tokens": 1000,
+    "median_confirmation_speedup": 1.8515941651996162,
+    "memory_gate": {
+      "ar_peak_memory_gib": 17.040895924903452,
+      "lte_ar_plus_6_gib": true,
+      "lte_ar_times_1_18": true,
+      "mtp_peak_memory_gib": 17.566339950077236
+    },
+    "mtp_by_block_size": {
+      "6": {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.410899550653994,
+        "block_size": 6,
+        "cache_memory_gib": 67.3336522737518,
+        "decode_s": 22.49344370800827,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566339950077236,
+        "prefill_s": 0.2826114170020446,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.8515941651996162,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.015003502820036374,
+          "draft": 1.5297440800641198,
+          "rollback": 0.00013703109289053828,
+          "target_distribution": 20.73823187011294,
+          "target_hidden": 0.0,
+          "verify": 0.1980874161090469
+        },
+        "tok_s": 44.457398919489286,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    },
+    "passed": false,
+    "prompt_id": "flappy_html5_canvas_game",
+    "prompt_tokens": 119,
+    "sampler": {
+      "seed": 0,
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    }
+  },
+  "status": "failed"
+}

benchmarks/flappy1000-targetq4-assistantq6-confirm2-block6.json

@@ -0,0 +1,1183 @@
+{
+  "arch_id": "gemma4-assistant-mtp",
+  "artifacts": {
+    "assistant_dtype": null,
+    "assistant_format": "q6-g64-affine",
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "assistant_quantization": {
+      "bits": 6,
+      "group_size": 64,
+      "mode": "affine"
+    },
+    "disk_ok": true,
+    "min_free_gib": 220.0,
+    "observed_free_gib": 704.1539611816406,
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+    "target_quantization": {
+      "bits": 4,
+      "format": "mlx-flat4-g64",
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "backend": "gemma4_assistant",
+  "benchmark": {
+    "draft_block_sizes": [
+      6
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "max_mode": true,
+    "max_tokens": 1000,
+    "profile": "sustained",
+    "prompt_suite": "mtplx/benchmarks/prompts/flappy.jsonl",
+    "reasoning": "off",
+    "sampler_source": {
+      "do_sample": true,
+      "local_reference": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx/generation_config.json",
+      "name": "official Gemma 4 generation_config.json",
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    },
+    "seed": 0,
+    "temperature": 1.0,
+    "top_k": 64,
+    "top_p": 0.95
+  },
+  "blockers": [
+    "best speedup 1.923x is below 2.000x",
+    "median confirmation speedup 1.923x is below 2.000x"
+  ],
+  "can_run_now": true,
+  "gates": {
+    "generated_tokens": 1000,
+    "longer_lengths_blocked_until_160_passes": true,
+    "median_of_3_min_speedup_vs_ar": 2.0,
+    "min_speedup_vs_ar": 2.0,
+    "mtp_peak_memory_lte_ar_multiplier": 1.18,
+    "mtp_peak_memory_lte_ar_plus_gib": 6
+  },
+  "official_sources": {
+    "assistant": "google/gemma-4-31B-it-assistant",
+    "assistant_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204",
+    "target": "google/gemma-4-31B-it",
+    "target_revision": "145dc2508c480a64b47242f160d286cff94a2343"
+  },
+  "pair": {
+    "assistant_exists": true,
+    "assistant_inspection": {
+      "architecture": "Gemma4AssistantForCausalLM",
+      "architecture_recognized": true,
+      "backbone_hidden_size": 5376,
+      "compatibility": {
+        "arch_id": "gemma4-assistant-mtp",
+        "can_run": false,
+        "exit_code": 3,
+        "message": "Official-style Gemma 4 31B assistant artifact recognized. This is an assistant-backed MTP pair, not a standalone target; MTPLX scaffold is present but QA and the 160-token speed/memory gate are still pending.",
+        "mtp_supported": "recognized",
+        "recognized": true,
+        "recommended_backend": "gemma4_assistant",
+        "recommended_profile": "performance-cold",
+        "runtime_compatibility": "assistant-pair-qa-pending",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "architecture-scaffolded-qa-pending",
+        "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+        "supported": false,
+        "tier": "architecture-compatible-but-unverified",
+        "unsafe_force_required": false,
+        "unverified_model": true
+      },
+      "config_exists": true,
+      "hidden_size": 1024,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+      "model_files": [
+        "model.safetensors"
+      ],
+      "model_type": "gemma4_assistant",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": "gemma4-assistant-mtp",
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "recognized",
+      "num_hidden_layers": 4,
+      "num_kv_shared_layers": 4,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 6,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": "gemma4_assistant",
+      "recommended_profile": "performance-cold",
+      "runtime_compatibility": "assistant-pair-qa-pending",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "architecture-scaffolded-qa-pending",
+      "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+      "unverified_model": true,
+      "use_ordered_embeddings": false,
+      "vocab_size": 262144
+    },
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "pair_error": null,
+    "pair_valid": true,
+    "target_exists": true,
+    "target_inspection": {
+      "architecture": "Gemma4ForConditionalGeneration",
+      "architecture_recognized": false,
+      "backbone_hidden_size": null,
+      "compatibility": {
+        "arch_id": null,
+        "can_run": false,
+        "exit_code": 2,
+        "message": "Model has no MTP head. MTPLX requires an MTP-equipped model.",
+        "mtp_supported": "no",
+        "recognized": false,
+        "recommended_backend": null,
+        "recommended_profile": null,
+        "runtime_compatibility": "unsupported",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "unsupported",
+        "support_notes": null,
+        "supported": false,
+        "tier": "no-MTP",
+        "unsafe_force_required": false,
+        "unverified_model": false
+      },
+      "config_exists": true,
+      "hidden_size": 5376,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+      "model_files": [
+        "model-00001-of-00004.safetensors",
+        "model-00002-of-00004.safetensors",
+        "model-00003-of-00004.safetensors",
+        "model-00004-of-00004.safetensors"
+      ],
+      "model_type": "gemma4_text",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": null,
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "no",
+      "num_hidden_layers": 60,
+      "num_kv_shared_layers": 0,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 4,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": null,
+      "recommended_profile": null,
+      "runtime_compatibility": "unsupported",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "unsupported",
+      "support_notes": null,
+      "unverified_model": false,
+      "use_ordered_embeddings": null,
+      "vocab_size": 262144
+    },
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx"
+  },
+  "planned_commands": {
+    "assistant_bf16_snapshot": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it-assistant', revision='cffbbd2cea41ea56a0fa5b0487e0d445121fd204', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx')\"",
+    "gate": "mtplx bench gemma-mtp --target-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --assistant-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx --profile sustained --max --prompt-suite mtplx/benchmarks/prompts/flappy.jsonl --max-tokens 1000 --temperature 1.0 --top-p 0.95 --top-k 64 --seed 0 --reasoning off --draft-block-sizes 6  --json --output outputs/gemma4/flappy1000-q6assistant-pure-confirm2-block6.json",
+    "target_flat4_g64": "uv run python -m mlx_lm.convert --hf-path /Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48 --mlx-path /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --quantize --q-bits 4 --q-group-size 64 --q-mode affine",
+    "target_revision_download": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it', revision='145dc2508c480a64b47242f160d286cff94a2343', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48')\""
+  },
+  "qa_pending": true,
+  "results": {
+    "ar": {
+      "active_memory_gib": 16.96542397607118,
+      "cache_memory_gib": 1.9759961115196347,
+      "decode_s": 43.638336165997316,
+      "generated_tokens": 1000,
+      "mode": "ar",
+      "peak_memory_gib": 17.040895924903452,
+      "prefill_s": 0.7599931250006193,
+      "tok_s": 22.915630792981354,
+      "token_preview": [
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761
+      ]
+    },
+    "ar_confirmation": [
+      {
+        "active_memory_gib": 16.96542397607118,
+        "cache_memory_gib": 1.9759961115196347,
+        "decode_s": 43.638336165997316,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 17.040895924903452,
+        "prefill_s": 0.7599931250006193,
+        "tok_s": 22.915630792981354,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      }
+    ],
+    "best_block_confirmation": [
+      {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.410899550653994,
+        "block_size": 6,
+        "cache_memory_gib": 67.3336522737518,
+        "decode_s": 22.690279750007903,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566339950077236,
+        "prefill_s": 0.25343483399774414,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.9232171946219445,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.014734539974597283,
+          "draft": 1.5304279569390928,
+          "rollback": 0.00014716008445248008,
+          "target_distribution": 20.938332714038552,
+          "target_hidden": 0.0,
+          "verify": 0.19443316804245114
+        },
+        "tok_s": 44.071735166669846,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    ],
+    "best_block_size": 6,
+    "best_speedup": 1.9232171946219445,
+    "blockers": [
+      "best speedup 1.923x is below 2.000x",
+      "median confirmation speedup 1.923x is below 2.000x"
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "fan_restore": {
+      "after": {
+        "actual_max_rpm": 6341,
+        "actual_min_rpm": 5980,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 5980,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5980,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5490
+            },
+            "rpm": 5980,
+            "target_rpm": 5490
+          },
+          {
+            "actual_rpm": 6341,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 6341,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5929
+            },
+            "rpm": 6341,
+            "target_rpm": 5929
+          }
+        ],
+        "max_rpm": 6341,
+        "min_rpm": 5980,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5980,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5490\n    },\n    {\n      \"actual_rpm\" : 6341,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5929\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 77.2,\n    \"TCHP\" : 64.2,\n    \"TCMb\" : 92.3,\n    \"TG0B\" : 32.5,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.5,\n    \"TH0x\" : 41.2,\n    \"TMVR\" : 69.4,\n    \"TPDX\" : 70.4,\n    \"TRDX\" : 83.8,\n    \"TS0P\" : 76.1,\n    \"Tg0j\" : 80.6,\n    \"Tm08\" : 76.7,\n    \"Tp04\" : 77.7,\n    \"Tp08\" : 77.3,\n    \"Tp0C\" : 78.1,\n    \"Tp0G\" : 78.5,\n    \"Tp0X\" : 78.3\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5980,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5490\n    },\n    {\n      \"actual_rpm\" : 6341,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5929\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.2,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 77.2,\n    \"TCHP\" : 64.2,\n    \"TCMb\" : 92.3,\n    \"TG0B\" : 32.5,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.5,\n    \"TH0x\" : 41.2,\n    \"TMVR\" : 69.4,\n    \"TPDX\" : 70.4,\n    \"TRDX\" : 83.8,\n    \"TS0P\" : 76.1,\n    \"Tg0j\" : 80.6,\n    \"Tm08\" : 76.7,\n    \"Tp04\" : 77.7,\n    \"Tp08\" : 77.3,\n    \"Tp0C\" : 78.1,\n    \"Tp0G\" : 78.5,\n    \"Tp0X\" : 78.3\n  }\n}"
+          }
+        },
+        "target_max_rpm": 5929,
+        "target_min_rpm": 5490
+      },
+      "message": "fan profile restored",
+      "ok": true,
+      "profile": "silent",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "auto"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "No matching processes were found",
+            "stdout": "Fans reset to Apple defaults"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "auto"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "silent"
+      }
+    },
+    "fanmax": {
+      "after": {
+        "actual_max_rpm": 7385,
+        "actual_min_rpm": 7332,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 7332,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7332,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7332,
+            "target_rpm": 7826
+          },
+          {
+            "actual_rpm": 7385,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7385,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7385,
+            "target_rpm": 7826
+          }
+        ],
+        "max_rpm": 7385,
+        "min_rpm": 7332,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7332,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7385,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 82.5,\n    \"TCHP\" : 60.5,\n    \"TCMb\" : 86.9,\n    \"TG0B\" : 32.6,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.6,\n    \"TH0x\" : 40,\n    \"TMVR\" : 59.5,\n    \"TPDX\" : 65.2,\n    \"TRDX\" : 67.7,\n    \"TS0P\" : 66.7,\n    \"Tg0j\" : 67.2,\n    \"Tm08\" : 67.9,\n    \"Tp04\" : 75.1,\n    \"Tp08\" : 74.2,\n    \"Tp0C\" : 77.5,\n    \"Tp0G\" : 76.2,\n    \"Tp0X\" : 75.2\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7332,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7385,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 82.5,\n    \"TCHP\" : 60.5,\n    \"TCMb\" : 86.9,\n    \"TG0B\" : 32.6,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.6,\n    \"TH0x\" : 40,\n    \"TMVR\" : 59.5,\n    \"TPDX\" : 65.2,\n    \"TRDX\" : 67.7,\n    \"TS0P\" : 66.7,\n    \"Tg0j\" : 67.2,\n    \"Tm08\" : 67.9,\n    \"Tp04\" : 75.1,\n    \"Tp08\" : 74.2,\n    \"Tp0C\" : 77.5,\n    \"Tp0G\" : 76.2,\n    \"Tp0X\" : 75.2\n  }\n}"
+          }
+        },
+        "target_max_rpm": 7826,
+        "target_min_rpm": 7826
+      },
+      "baseline": {
+        "actual_max_rpm": 5367,
+        "actual_min_rpm": 4968,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 4968,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 4968,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 4976
+            },
+            "rpm": 4968,
+            "target_rpm": 4976
+          },
+          {
+            "actual_rpm": 5367,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5367,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5374
+            },
+            "rpm": 5367,
+            "target_rpm": 5374
+          }
+        ],
+        "max_rpm": 5367,
+        "min_rpm": 4968,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 4968,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4976\n    },\n    {\n      \"actual_rpm\" : 5367,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5374\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 68.2,\n    \"TCHP\" : 60.5,\n    \"TCMb\" : 75.8,\n    \"TG0B\" : 32.6,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.6,\n    \"TH0x\" : 39.6,\n    \"TMVR\" : 59.7,\n    \"TPDX\" : 65.7,\n    \"TRDX\" : 67.6,\n    \"TS0P\" : 67.4,\n    \"Tg0j\" : 67.2,\n    \"Tm08\" : 68.6,\n    \"Tp04\" : 76.5,\n    \"Tp08\" : 75.4,\n    \"Tp0C\" : 79.5,\n    \"Tp0G\" : 77.4,\n    \"Tp0X\" : 76.6\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 4968,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 4976\n    },\n    {\n      \"actual_rpm\" : 5367,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5374\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.1,\n    \"TB0T\" : 32.6,\n    \"TCDX\" : 68.2,\n    \"TCHP\" : 60.5,\n    \"TCMb\" : 75.8,\n    \"TG0B\" : 32.6,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.6,\n    \"TH0x\" : 39.6,\n    \"TMVR\" : 59.7,\n    \"TPDX\" : 65.7,\n    \"TRDX\" : 67.6,\n    \"TS0P\" : 67.4,\n    \"Tg0j\" : 67.2,\n    \"Tm08\" : 68.6,\n    \"Tp04\" : 76.5,\n    \"Tp08\" : 75.4,\n    \"Tp0C\" : 79.5,\n    \"Tp0G\" : 77.4,\n    \"Tp0X\" : 76.6\n  }\n}"
+          }
+        },
+        "target_max_rpm": 5374,
+        "target_min_rpm": 4976
+      },
+      "message": "fans ramped to max (actual 7332-7385 RPM; target 7826 RPM)",
+      "ok": true,
+      "profile": "max",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "max"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "Fan 0: 4968 RPM \u2192 max (7826 RPM)\nFan 1: 5367 RPM \u2192 max (7826 RPM)"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "max"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "max"
+      }
+    },
+    "max_tokens": 1000,
+    "median_confirmation_speedup": 1.9232171946219445,
+    "memory_gate": {
+      "ar_peak_memory_gib": 17.040895924903452,
+      "lte_ar_plus_6_gib": true,
+      "lte_ar_times_1_18": true,
+      "mtp_peak_memory_gib": 17.566339950077236
+    },
+    "mtp_by_block_size": {
+      "6": {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.410899550653994,
+        "block_size": 6,
+        "cache_memory_gib": 67.3336522737518,
+        "decode_s": 22.690279750007903,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566339950077236,
+        "prefill_s": 0.25343483399774414,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.9232171946219445,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.014734539974597283,
+          "draft": 1.5304279569390928,
+          "rollback": 0.00014716008445248008,
+          "target_distribution": 20.938332714038552,
+          "target_hidden": 0.0,
+          "verify": 0.19443316804245114
+        },
+        "tok_s": 44.071735166669846,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    },
+    "passed": false,
+    "prompt_id": "flappy_html5_canvas_game",
+    "prompt_tokens": 119,
+    "sampler": {
+      "seed": 0,
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    }
+  },
+  "status": "failed"
+}

benchmarks/flappy1000-targetq4-assistantq6-sweep.json

@@ -0,0 +1,1702 @@
+{
+  "arch_id": "gemma4-assistant-mtp",
+  "artifacts": {
+    "assistant_dtype": null,
+    "assistant_format": "q6-g64-affine",
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "assistant_quantization": {
+      "bits": 6,
+      "group_size": 64,
+      "mode": "affine"
+    },
+    "disk_ok": true,
+    "min_free_gib": 220.0,
+    "observed_free_gib": 704.0457382202148,
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+    "target_quantization": {
+      "bits": 4,
+      "format": "mlx-flat4-g64",
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "backend": "gemma4_assistant",
+  "benchmark": {
+    "draft_block_sizes": [
+      3,
+      4,
+      5,
+      6
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "max_mode": true,
+    "max_tokens": 1000,
+    "profile": "sustained",
+    "prompt_suite": "mtplx/benchmarks/prompts/flappy.jsonl",
+    "reasoning": "off",
+    "sampler_source": {
+      "do_sample": true,
+      "local_reference": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx/generation_config.json",
+      "name": "official Gemma 4 generation_config.json",
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    },
+    "seed": 0,
+    "temperature": 1.0,
+    "top_k": 64,
+    "top_p": 0.95
+  },
+  "blockers": [
+    "best speedup 1.995x is below 2.000x",
+    "median confirmation speedup 1.995x is below 2.000x"
+  ],
+  "can_run_now": true,
+  "gates": {
+    "generated_tokens": 1000,
+    "longer_lengths_blocked_until_160_passes": true,
+    "median_of_3_min_speedup_vs_ar": 2.0,
+    "min_speedup_vs_ar": 2.0,
+    "mtp_peak_memory_lte_ar_multiplier": 1.18,
+    "mtp_peak_memory_lte_ar_plus_gib": 6
+  },
+  "official_sources": {
+    "assistant": "google/gemma-4-31B-it-assistant",
+    "assistant_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204",
+    "target": "google/gemma-4-31B-it",
+    "target_revision": "145dc2508c480a64b47242f160d286cff94a2343"
+  },
+  "pair": {
+    "assistant_exists": true,
+    "assistant_inspection": {
+      "architecture": "Gemma4AssistantForCausalLM",
+      "architecture_recognized": true,
+      "backbone_hidden_size": 5376,
+      "compatibility": {
+        "arch_id": "gemma4-assistant-mtp",
+        "can_run": false,
+        "exit_code": 3,
+        "message": "Official-style Gemma 4 31B assistant artifact recognized. This is an assistant-backed MTP pair, not a standalone target; MTPLX scaffold is present but QA and the 160-token speed/memory gate are still pending.",
+        "mtp_supported": "recognized",
+        "recognized": true,
+        "recommended_backend": "gemma4_assistant",
+        "recommended_profile": "performance-cold",
+        "runtime_compatibility": "assistant-pair-qa-pending",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "architecture-scaffolded-qa-pending",
+        "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+        "supported": false,
+        "tier": "architecture-compatible-but-unverified",
+        "unsafe_force_required": false,
+        "unverified_model": true
+      },
+      "config_exists": true,
+      "hidden_size": 1024,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+      "model_files": [
+        "model.safetensors"
+      ],
+      "model_type": "gemma4_assistant",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": "gemma4-assistant-mtp",
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "recognized",
+      "num_hidden_layers": 4,
+      "num_kv_shared_layers": 4,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 6,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": "gemma4_assistant",
+      "recommended_profile": "performance-cold",
+      "runtime_compatibility": "assistant-pair-qa-pending",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "architecture-scaffolded-qa-pending",
+      "support_notes": "Assistant-backed scaffold for the official dense Gemma 4 31B pair. It remains QA-pending and is not a public runnable backend until 160-token exactness, speed, and memory gates pass.",
+      "unverified_model": true,
+      "use_ordered_embeddings": false,
+      "vocab_size": 262144
+    },
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx",
+    "pair_error": null,
+    "pair_valid": true,
+    "target_exists": true,
+    "target_inspection": {
+      "architecture": "Gemma4ForConditionalGeneration",
+      "architecture_recognized": false,
+      "backbone_hidden_size": null,
+      "compatibility": {
+        "arch_id": null,
+        "can_run": false,
+        "exit_code": 2,
+        "message": "Model has no MTP head. MTPLX requires an MTP-equipped model.",
+        "mtp_supported": "no",
+        "recognized": false,
+        "recommended_backend": null,
+        "recommended_profile": null,
+        "runtime_compatibility": "unsupported",
+        "runtime_contract": null,
+        "runtime_contract_error": null,
+        "runtime_contract_path": null,
+        "support_level": "unsupported",
+        "support_notes": null,
+        "supported": false,
+        "tier": "no-MTP",
+        "unsafe_force_required": false,
+        "unverified_model": false
+      },
+      "config_exists": true,
+      "hidden_size": 5376,
+      "layer_types": [
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "sliding_attention",
+        "full_attention"
+      ],
+      "model_dir": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx",
+      "model_files": [
+        "model-00001-of-00004.safetensors",
+        "model-00002-of-00004.safetensors",
+        "model-00003-of-00004.safetensors",
+        "model-00004-of-00004.safetensors"
+      ],
+      "model_type": "gemma4_text",
+      "mtp": {
+        "exists": false,
+        "expected_tensor_count": 15,
+        "extra_keys": [],
+        "metadata_only": true,
+        "missing_expected_keys": [],
+        "mtp_file": "model.safetensors.index.json::embedded",
+        "passes_tensor_gate": false,
+        "sidecar_format": "bf16",
+        "tensor_count": 0,
+        "tensors": []
+      },
+      "mtp_arch": null,
+      "mtp_num_hidden_layers": 0,
+      "mtp_pattern": null,
+      "mtp_supported": "no",
+      "num_hidden_layers": 60,
+      "num_kv_shared_layers": 0,
+      "passes_primary_gate": false,
+      "quantization": {
+        "bits": 4,
+        "group_size": 64,
+        "mode": "affine"
+      },
+      "recommended_backend": null,
+      "recommended_profile": null,
+      "runtime_compatibility": "unsupported",
+      "runtime_contract_path": null,
+      "sidecars": {
+        "preprocessor_config.json": false,
+        "processor_config.json": false,
+        "video_preprocessor_config.json": false
+      },
+      "source": "local",
+      "support_level": "unsupported",
+      "support_notes": null,
+      "unverified_model": false,
+      "use_ordered_embeddings": null,
+      "vocab_size": 262144
+    },
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx"
+  },
+  "planned_commands": {
+    "assistant_bf16_snapshot": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it-assistant', revision='cffbbd2cea41ea56a0fa5b0487e0d445121fd204', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx')\"",
+    "gate": "mtplx bench gemma-mtp --target-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --assistant-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q6-g64-mlx --profile sustained --max --prompt-suite mtplx/benchmarks/prompts/flappy.jsonl --max-tokens 1000 --temperature 1.0 --top-p 0.95 --top-k 64 --seed 0 --reasoning off --draft-block-sizes 3,4,5,6  --json --output outputs/gemma4/flappy1000-q6assistant-pure-sweep.json",
+    "target_flat4_g64": "uv run python -m mlx_lm.convert --hf-path /Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48 --mlx-path /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-flat4-g64-mlx --quantize --q-bits 4 --q-group-size 64 --q-mode affine",
+    "target_revision_download": "uv run python -c \"from huggingface_hub import snapshot_download; snapshot_download(repo_id='google/gemma-4-31B-it', revision='145dc2508c480a64b47242f160d286cff94a2343', repo_type='model', local_dir='/Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48')\""
+  },
+  "qa_pending": true,
+  "results": {
+    "ar": {
+      "active_memory_gib": 16.96541445143521,
+      "cache_memory_gib": 1.9738727556541562,
+      "decode_s": 45.80454516700411,
+      "generated_tokens": 1000,
+      "mode": "ar",
+      "peak_memory_gib": 17.040886400267482,
+      "prefill_s": 1.318873333002557,
+      "tok_s": 21.831894550071045,
+      "token_preview": [
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761
+      ]
+    },
+    "ar_confirmation": [
+      {
+        "active_memory_gib": 16.96541445143521,
+        "cache_memory_gib": 1.9738727556541562,
+        "decode_s": 45.80454516700411,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 17.040886400267482,
+        "prefill_s": 1.318873333002557,
+        "tok_s": 21.831894550071045,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      }
+    ],
+    "best_block_confirmation": [
+      {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.41092054359615,
+        "block_size": 6,
+        "cache_memory_gib": 67.33365368191153,
+        "decode_s": 22.95747512500384,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566345684230328,
+        "prefill_s": 0.2074422090081498,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.9951908873949589,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.011550506053026766,
+          "draft": 1.5224984570086235,
+          "rollback": 0.0001419630425516516,
+          "target_distribution": 21.21943625298445,
+          "target_hidden": 0.0,
+          "verify": 0.19109521005884744
+        },
+        "tok_s": 43.558797060869416,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    ],
+    "best_block_size": 6,
+    "best_speedup": 1.9951908873949589,
+    "blockers": [
+      "best speedup 1.995x is below 2.000x",
+      "median confirmation speedup 1.995x is below 2.000x"
+    ],
+    "draft_sampler": {
+      "exactness_note": "Assistant q may differ from target p; MTPLX remains exact because acceptance uses p/q and rejection samples the residual distribution.",
+      "inherits_target_sampler": true,
+      "temperature": null,
+      "top_k": null,
+      "top_p": null
+    },
+    "fan_restore": {
+      "after": {
+        "actual_max_rpm": 6643,
+        "actual_min_rpm": 6265,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 6265,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 6265,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5768
+            },
+            "rpm": 6265,
+            "target_rpm": 5768
+          },
+          {
+            "actual_rpm": 6643,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 6643,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 6229
+            },
+            "rpm": 6643,
+            "target_rpm": 6229
+          }
+        ],
+        "max_rpm": 6643,
+        "min_rpm": 6265,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 6265,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5768\n    },\n    {\n      \"actual_rpm\" : 6643,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6229\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.3,\n    \"TCDX\" : 73.6,\n    \"TCHP\" : 61.1,\n    \"TCMb\" : 88.7,\n    \"TG0B\" : 32.4,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.4,\n    \"TH0x\" : 40.9,\n    \"TMVR\" : 66.4,\n    \"TPDX\" : 68.6,\n    \"TRDX\" : 81.9,\n    \"TS0P\" : 72.8,\n    \"Tg0j\" : 77,\n    \"Tm08\" : 73.7,\n    \"Tp04\" : 74.2,\n    \"Tp08\" : 73.9,\n    \"Tp0C\" : 74.8,\n    \"Tp0G\" : 75.2,\n    \"Tp0X\" : 75\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 6265,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5768\n    },\n    {\n      \"actual_rpm\" : 6643,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6229\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.3,\n    \"TCDX\" : 73.6,\n    \"TCHP\" : 61.1,\n    \"TCMb\" : 88.7,\n    \"TG0B\" : 32.4,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.4,\n    \"TH0x\" : 40.9,\n    \"TMVR\" : 66.4,\n    \"TPDX\" : 68.6,\n    \"TRDX\" : 81.9,\n    \"TS0P\" : 72.8,\n    \"Tg0j\" : 77,\n    \"Tm08\" : 73.7,\n    \"Tp04\" : 74.2,\n    \"Tp08\" : 73.9,\n    \"Tp0C\" : 74.8,\n    \"Tp0G\" : 75.2,\n    \"Tp0X\" : 75\n  }\n}"
+          }
+        },
+        "target_max_rpm": 6229,
+        "target_min_rpm": 5768
+      },
+      "message": "fan profile restored",
+      "ok": true,
+      "profile": "silent",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "auto"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "No matching processes were found",
+            "stdout": "Fans reset to Apple defaults"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "auto"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "silent"
+      }
+    },
+    "fanmax": {
+      "after": {
+        "actual_max_rpm": 7477,
+        "actual_min_rpm": 7385,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 7385,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7385,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7385,
+            "target_rpm": 7826
+          },
+          {
+            "actual_rpm": 7477,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7477,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7477,
+            "target_rpm": 7826
+          }
+        ],
+        "max_rpm": 7477,
+        "min_rpm": 7385,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7385,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7477,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.8,\n    \"TCDX\" : 64.5,\n    \"TCHP\" : 56,\n    \"TCMb\" : 71.5,\n    \"TG0B\" : 32.8,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.8,\n    \"TH0x\" : 38.8,\n    \"TMVR\" : 55.8,\n    \"TPDX\" : 60.8,\n    \"TRDX\" : 60.9,\n    \"TS0P\" : 61.7,\n    \"Tg0j\" : 61,\n    \"Tm08\" : 61.6,\n    \"Tp04\" : 64.6,\n    \"Tp08\" : 64,\n    \"Tp0C\" : 65.5,\n    \"Tp0G\" : 65.3,\n    \"Tp0X\" : 64\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7385,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7477,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.8,\n    \"TCDX\" : 64.5,\n    \"TCHP\" : 56,\n    \"TCMb\" : 71.5,\n    \"TG0B\" : 32.8,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.8,\n    \"TH0x\" : 38.8,\n    \"TMVR\" : 55.8,\n    \"TPDX\" : 60.8,\n    \"TRDX\" : 60.9,\n    \"TS0P\" : 61.7,\n    \"Tg0j\" : 61,\n    \"Tm08\" : 61.6,\n    \"Tp04\" : 64.6,\n    \"Tp08\" : 64,\n    \"Tp0C\" : 65.5,\n    \"Tp0G\" : 65.3,\n    \"Tp0X\" : 64\n  }\n}"
+          }
+        },
+        "target_max_rpm": 7826,
+        "target_min_rpm": 7826
+      },
+      "baseline": {
+        "actual_max_rpm": 6286,
+        "actual_min_rpm": 5825,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 5825,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5825,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5822
+            },
+            "rpm": 5825,
+            "target_rpm": 5822
+          },
+          {
+            "actual_rpm": 6286,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 6286,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 6287
+            },
+            "rpm": 6286,
+            "target_rpm": 6287
+          }
+        ],
+        "max_rpm": 6286,
+        "min_rpm": 5825,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5825,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5822\n    },\n    {\n      \"actual_rpm\" : 6286,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6287\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.8,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 56.4,\n    \"TCMb\" : 73.3,\n    \"TG0B\" : 32.8,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.8,\n    \"TH0x\" : 38.8,\n    \"TMVR\" : 56.4,\n    \"TPDX\" : 61.2,\n    \"TRDX\" : 61.4,\n    \"TS0P\" : 62.7,\n    \"Tg0j\" : 61.3,\n    \"Tm08\" : 61.7,\n    \"Tp04\" : 63.4,\n    \"Tp08\" : 62.9,\n    \"Tp0C\" : 63.8,\n    \"Tp0G\" : 64,\n    \"Tp0X\" : 63.1\n  }\n}"
+            }
+          ],
+          "detection": {
+            "available": true,
+            "clock_anchor_enabled": false,
+            "clock_anchor_policy": "explicit experimental only; never used for product claims",
+            "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+            "selected": {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            },
+            "tools": [
+              {
+                "kind": "thermalforge",
+                "path": "/Users/youssof/.mtplx/bin/thermalforge",
+                "version": {
+                  "command": [
+                    "/Users/youssof/.mtplx/bin/thermalforge",
+                    "--version"
+                  ],
+                  "ok": true,
+                  "returncode": 0,
+                  "stderr": "",
+                  "stdout": "0.1.0"
+                }
+              }
+            ]
+          },
+          "ok": true,
+          "status": {
+            "command": [
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "status"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5825,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5822\n    },\n    {\n      \"actual_rpm\" : 6286,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6287\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.7,\n    \"TB0T\" : 32.8,\n    \"TCDX\" : 63.9,\n    \"TCHP\" : 56.4,\n    \"TCMb\" : 73.3,\n    \"TG0B\" : 32.8,\n    \"TG0H\" : 32,\n    \"TG0V\" : 32.8,\n    \"TH0x\" : 38.8,\n    \"TMVR\" : 56.4,\n    \"TPDX\" : 61.2,\n    \"TRDX\" : 61.4,\n    \"TS0P\" : 62.7,\n    \"Tg0j\" : 61.3,\n    \"Tm08\" : 61.7,\n    \"Tp04\" : 63.4,\n    \"Tp08\" : 62.9,\n    \"Tp0C\" : 63.8,\n    \"Tp0G\" : 64,\n    \"Tp0X\" : 63.1\n  }\n}"
+          }
+        },
+        "target_max_rpm": 6287,
+        "target_min_rpm": 5822
+      },
+      "message": "fans ramped to max (actual 7385-7477 RPM; target 7826 RPM)",
+      "ok": true,
+      "profile": "max",
+      "set_result": {
+        "attempts": [
+          {
+            "command": [
+              "sudo",
+              "-n",
+              "/Users/youssof/.mtplx/bin/thermalforge",
+              "max"
+            ],
+            "ok": true,
+            "returncode": 0,
+            "stderr": "",
+            "stdout": "Fan 0: 5825 RPM \u2192 max (7826 RPM)\nFan 1: 6286 RPM \u2192 max (7826 RPM)"
+          }
+        ],
+        "command": [
+          "sudo",
+          "-n",
+          "/Users/youssof/.mtplx/bin/thermalforge",
+          "max"
+        ],
+        "detection": {
+          "available": true,
+          "clock_anchor_enabled": false,
+          "clock_anchor_policy": "explicit experimental only; never used for product claims",
+          "instructions": "Install ThermalForge and ensure the thermalforge CLI is on PATH.",
+          "selected": {
+            "kind": "thermalforge",
+            "path": "/Users/youssof/.mtplx/bin/thermalforge",
+            "version": {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "--version"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "0.1.0"
+            }
+          },
+          "tools": [
+            {
+              "kind": "thermalforge",
+              "path": "/Users/youssof/.mtplx/bin/thermalforge",
+              "version": {
+                "command": [
+                  "/Users/youssof/.mtplx/bin/thermalforge",
+                  "--version"
+                ],
+                "ok": true,
+                "returncode": 0,
+                "stderr": "",
+                "stdout": "0.1.0"
+              }
+            }
+          ]
+        },
+        "dry_run": false,
+        "ok": true,
+        "profile": "max"
+      }
+    },
+    "max_tokens": 1000,
+    "median_confirmation_speedup": 1.9951908873949589,
+    "memory_gate": {
+      "ar_peak_memory_gib": 17.040886400267482,
+      "lte_ar_plus_6_gib": true,
+      "lte_ar_times_1_18": true,
+      "mtp_peak_memory_gib": 17.566345684230328
+    },
+    "mtp_by_block_size": {
+      "3": {
+        "acceptance": 0.991044776119403,
+        "accepted_drafts": 664,
+        "active_memory_gib": 17.409974498674273,
+        "block_size": 3,
+        "cache_memory_gib": 96.58854798320681,
+        "decode_s": 37.873455042004935,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 670,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.5211139684543,
+        "prefill_s": 0.3529969999945024,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.2094102615196556,
+        "target_distribution_modes": {
+          "batched_logits": 335
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.0206884491344681,
+          "draft": 1.4088243619044079,
+          "rollback": 0.0002345078537473455,
+          "target_distribution": 36.03245366179908,
+          "target_hidden": 0.0,
+          "verify": 0.3872644162038341
+        },
+        "tok_s": 26.403717297270966,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 335
+      },
+      "4": {
+        "acceptance": 0.9548969072164949,
+        "accepted_drafts": 741,
+        "active_memory_gib": 17.411866588518023,
+        "block_size": 4,
+        "cache_memory_gib": 96.58724461961538,
+        "decode_s": 36.01139437498932,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 776,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.544619735330343,
+        "prefill_s": 0.2180854579928564,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.2719458927371148,
+        "target_distribution_modes": {
+          "batched_logits": 259
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.01497975607344415,
+          "draft": 1.4912097119231476,
+          "rollback": 0.0002762930525932461,
+          "target_distribution": 34.18285454783472,
+          "target_hidden": 0.0,
+          "verify": 0.3010120859835297
+        },
+        "tok_s": 27.768988603632668,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          236761,
+          5715,
+          236789,
+          236751,
+          506,
+          3772,
+          236787
+        ],
+        "verify_calls": 259
+      },
+      "5": {
+        "acceptance": 0.8822197055492639,
+        "accepted_drafts": 779,
+        "active_memory_gib": 17.411210460588336,
+        "block_size": 5,
+        "cache_memory_gib": 82.11562378518283,
+        "decode_s": 24.697525832991232,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 883,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.551317367702723,
+        "prefill_s": 0.21940916699531954,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.8546208019678587,
+        "target_distribution_modes": {
+          "batched_logits": 221
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.013920457829954103,
+          "draft": 1.5722365041146986,
+          "rollback": 0.00047238002298399806,
+          "target_distribution": 22.840641582908574,
+          "target_hidden": 0.0,
+          "verify": 0.2542634050187189
+        },
+        "tok_s": 40.489885778930486,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          236761,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 221
+      },
+      "6": {
+        "acceptance": 0.9810874704491725,
+        "accepted_drafts": 830,
+        "active_memory_gib": 17.41092054359615,
+        "block_size": 6,
+        "cache_memory_gib": 67.33365368191153,
+        "decode_s": 22.95747512500384,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 846,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 17.566345684230328,
+        "prefill_s": 0.2074422090081498,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 1.9951908873949589,
+        "target_distribution_modes": {
+          "batched_logits": 170
+        },
+        "target_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "telemetry": {
+          "ar_dense_fallback_calls": 0,
+          "decode_dense_fallback_calls": 0,
+          "dense_fallback_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "events": [],
+          "paged_active_array_calls_by_phase": {
+            "ar_decode": 0,
+            "decode_verify": 0,
+            "postcommit": 0,
+            "prefill": 0,
+            "unknown": 0
+          },
+          "paged_attention_bailouts_by_phase_reason": {
+            "ar_decode": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "decode_verify": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "postcommit": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "prefill": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            },
+            "unknown": {
+              "batch_not_1": 0,
+              "block_size_mismatch": 0,
+              "blocks_invalid": 0,
+              "dtype_unsupported": 0,
+              "empty_cache": 0,
+              "head_dim_unsupported": 0,
+              "kernel_unavailable": 0,
+              "offset_invalid": 0,
+              "partitioned_invalid_output": 0,
+              "partitioned_unavailable": 0,
+              "q_len_gt_max": 0,
+              "q_len_invalid": 0,
+              "turboquant_unsupported": 0,
+              "unknown": 0,
+              "unsupported_mask": 0
+            }
+          },
+          "paged_attention_large_q_path": {
+            "dense_forbidden": 0,
+            "large_q_split_sdpa_fallback": 0,
+            "partitioned_paged": 0,
+            "tail_paged": 0,
+            "unknown": 0
+          },
+          "postcommit_dense_fallback_calls": 0,
+          "prefill_dense_fallback_calls": 0,
+          "trace_events": false
+        },
+        "timing_s": {
+          "accept": 0.011550506053026766,
+          "draft": 1.5224984570086235,
+          "rollback": 0.0001419630425516516,
+          "target_distribution": 21.21943625298445,
+          "target_hidden": 0.0,
+          "verify": 0.19109521005884744
+        },
+        "tok_s": 43.558797060869416,
+        "token_preview": [
+          9996,
+          625,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759,
+          759
+        ],
+        "verify_calls": 170
+      }
+    },
+    "passed": false,
+    "prompt_id": "flappy_html5_canvas_game",
+    "prompt_tokens": 119,
+    "sampler": {
+      "seed": 0,
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    }
+  },
+  "status": "failed"
+}

mtplx_pair.json

@@ -0,0 +1,52 @@
+{
+  "format_version": 1,
+  "name": "Gemma4 MTPLX Optimized Speed",
+  "variant": "speed",
+  "layout": {
+    "target": "target",
+    "assistant": "assistant"
+  },
+  "source": {
+    "target_repo": "google/gemma-4-31B-it",
+    "target_revision": "145dc2508c480a64b47242f160d286cff94a2343",
+    "assistant_repo": "google/gemma-4-31B-it-assistant",
+    "assistant_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204"
+  },
+  "target": {
+    "role": "verifier",
+    "model_type": "gemma4",
+    "quantization": {
+      "bits": 4,
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "assistant": {
+    "role": "drafter",
+    "model_type": "gemma4_assistant",
+    "quantization": {
+      "bits": 6,
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "benchmark": {
+    "prompt_suite": "flappy",
+    "max_tokens": 1000,
+    "temperature": 1.0,
+    "top_p": 0.95,
+    "top_k": 64,
+    "seed": 0,
+    "best_block_size": 6,
+    "acceptance": {
+      "accepted": 830,
+      "drafted": 846,
+      "ratio": 0.9810874704491725
+    },
+    "observed_mtp_tok_s": [
+      43.558797060869416,
+      44.457398919489286,
+      44.071735166669846
+    ]
+  }
+}

target/README.md

@@ -0,0 +1,7 @@
+---
+language: en
+tags:
+- mlx
+pipeline_tag: text-generation
+library_name: mlx
+---

target/config.json

@@ -0,0 +1,144 @@
+{
+    "architectures": [
+        "Gemma4ForConditionalGeneration"
+    ],
+    "audio_config": null,
+    "audio_token_id": 258881,
+    "boa_token_id": 256000,
+    "boi_token_id": 255999,
+    "dtype": "bfloat16",
+    "eoa_token_id": 258883,
+    "eoa_token_index": 258883,
+    "eoi_token_id": 258882,
+    "eos_token_id": [
+        1,
+        106,
+        50
+    ],
+    "image_token_id": 258880,
+    "initializer_range": 0.02,
+    "model_type": "gemma4",
+    "quantization": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine"
+    },
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attention_k_eq_v": true,
+        "bos_token_id": 2,
+        "dtype": "bfloat16",
+        "enable_moe_block": false,
+        "eos_token_id": 1,
+        "expert_intermediate_size": null,
+        "final_logit_softcapping": 30.0,
+        "global_head_dim": 512,
+        "head_dim": 256,
+        "hidden_activation": "gelu_pytorch_tanh",
+        "hidden_size": 5376,
+        "hidden_size_per_layer_input": 0,
+        "initializer_range": 0.02,
+        "intermediate_size": 21504,
+        "layer_types": [
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention"
+        ],
+        "max_position_embeddings": 262144,
+        "model_type": "gemma4_text",
+        "num_attention_heads": 32,
+        "num_experts": null,
+        "num_global_key_value_heads": 4,
+        "num_hidden_layers": 60,
+        "num_key_value_heads": 16,
+        "num_kv_shared_layers": 0,
+        "pad_token_id": 0,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "full_attention": {
+                "partial_rotary_factor": 0.25,
+                "rope_theta": 1000000.0,
+                "rope_type": "proportional"
+            },
+            "sliding_attention": {
+                "rope_theta": 10000.0,
+                "rope_type": "default"
+            }
+        },
+        "sliding_window": 1024,
+        "tie_word_embeddings": true,
+        "top_k_experts": null,
+        "use_bidirectional_attention": "vision",
+        "use_cache": true,
+        "use_double_wide_mlp": false,
+        "vocab_size": 262144,
+        "vocab_size_per_layer_input": 262144
+    },
+    "tie_word_embeddings": true,
+    "transformers_version": "5.5.0.dev0",
+    "video_token_id": 258884,
+    "vision_soft_tokens_per_image": 280
+}

target/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 2,
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106,
+    50
+  ],
+  "pad_token_id": 0,
+  "temperature": 1.0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.5.0.dev0"
+}

target/model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:988e2b1fd41d93b62b8c432f52f632c43b8cb7f86df4b957db36a3cc0dab40ca
+size 5366617512

target/model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a496a96fbd39cd11a9871f91d026013d91069dcac15f89ca861b93976f3857cf
+size 5361642573

target/model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afa555ff0e1bc458c5b08aeef1f4499dce63e2bfb5d3a2aac716e47c0a5672c1
+size 5367276094

target/model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bde02793a3d2ed3f29c3d1629ea9fbeb9c25720bc42db0eca9720abe094235a
+size 1173848301

target/mtplx_artifact.json

@@ -0,0 +1,18 @@
+{
+  "artifact_schema": "mtplx.gemma4.google.v1",
+  "config.json_sha256": "b2a7c25dc48e584d3debc824534a62cd73a801108b1cc7a15d32676195d7f03a",
+  "created_at_unix": 1778037794,
+  "generation_config.json_sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
+  "quantization": {
+    "bits": 4,
+    "format": "mlx-flat4-g64",
+    "group_size": 64,
+    "mode": "affine"
+  },
+  "role": "target",
+  "source_repo": "google/gemma-4-31B-it",
+  "source_revision": "145dc2508c480a64b47242f160d286cff94a2343",
+  "source_snapshot": "/Users/youssof/Documents/MTPLX/models/.sources/gemma-4-31B-it-145dc2508c48",
+  "tokenizer.json_sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
+  "tokenizer_config.json_sha256": "a284d1243b62be31faa9c13e1c28cece940c4abaa7bd9ad87b94f61b40687200"
+}

target/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f
+size 32169626

target/tokenizer_config.json

@@ -0,0 +1,96 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [
+    "<|video|>"
+  ],
+  "image_token": "<|image|>",
+  "is_local": true,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "processor_class": "Gemma4Processor",
+  "response_schema": {
+    "properties": {
+      "content": {
+        "type": "string"
+      },
+      "role": {
+        "const": "assistant"
+      },
+      "thinking": {
+        "type": "string"
+      },
+      "tool_calls": {
+        "items": {
+          "properties": {
+            "function": {
+              "properties": {
+                "arguments": {
+                  "additionalProperties": {},
+                  "type": "object",
+                  "x-parser": "gemma4-tool-call"
+                },
+                "name": {
+                  "type": "string"
+                }
+              },
+              "type": "object",
+              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})"
+            },
+            "type": {
+              "const": "function"
+            }
+          },
+          "type": "object"
+        },
+        "type": "array",
+        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>"
+      }
+    },
+    "type": "object",
+    "x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?P<content>(?:(?!\\<turn\\|\\>)(?!\\<\\|tool_response\\>).)+)?(?:\\<turn\\|\\>|\\<\\|tool_response\\>)?"
+  },
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>"
+}