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
+assistant/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+target/tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,107 @@
+---
+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 Quality
+
+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 Q8 affine group-size 64
+- `assistant/` - official Gemma 4 31B assistant drafter, MLX Q8 affine group-size 64
+
+Use this pair when target precision and high acceptance are 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: 8
+group_size: 64
+mode: affine
+```
+
+Assistant:
+
+```text
+bits: 8
+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: 833 / 835 = 99.76%
+speedup_vs_ar: 2.49x
+```
+
+Observed MTPLX throughput samples:
+
+```text
+34.22 tok/s
+32.88 tok/s
+33.12 tok/s
+```
+
+The bundled benchmark JSON file is in `benchmarks/`.
+
+## Notes
+
+This release is optimized for target precision and high acceptance. It is not the fastest absolute-TPS pair; for speed, use `Youssofal/Gemma4-MTPLX-Optimized-Speed`.
+
+Gemma 4 is released by Google under the Gemma 4 license terms linked above.

assistant/README.md

@@ -0,0 +1,7 @@
+---
+language: en
+tags:
+- mlx
+pipeline_tag: text-generation
+library_name: 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": 8,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 8,
+        "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 sha256:9503f95f4259bb000df3a89c5ab44ce94eed935d8e949926db8aa288ab1c3180
+size 498891309

assistant/model.safetensors.index.json

@@ -0,0 +1,102 @@
+{
+    "metadata": {
+        "total_size": 498881032,
+        "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": "Q8 affine G64 for all quantizable assistant modules, including tied embedding/LM-head path and projections.",
+  "quantization": {
+    "bits": 8,
+    "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-targetq8-assistantq8-sweep.json

@@ -0,0 +1,2104 @@
+{
+  "arch_id": "gemma4-assistant-mtp",
+  "artifacts": {
+    "assistant_dtype": null,
+    "assistant_format": "q8-g64-affine",
+    "assistant_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q8-g64-mlx",
+    "assistant_quantization": {
+      "bits": 8,
+      "group_size": 64,
+      "mode": "affine"
+    },
+    "disk_ok": true,
+    "min_free_gib": 220.0,
+    "observed_free_gib": 630.7444229125977,
+    "target_dtype": null,
+    "target_format": "q8-g64-affine",
+    "target_model": "/Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-q8-g64-mlx",
+    "target_quantization": {
+      "bits": 8,
+      "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-q8-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": [],
+  "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-q8-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": 8,
+        "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-q8-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-q8-g64-mlx",
+      "model_files": [
+        "model-00001-of-00007.safetensors",
+        "model-00002-of-00007.safetensors",
+        "model-00003-of-00007.safetensors",
+        "model-00004-of-00007.safetensors",
+        "model-00005-of-00007.safetensors",
+        "model-00006-of-00007.safetensors",
+        "model-00007-of-00007.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": 8,
+        "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-q8-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-q8-g64-mlx')\"",
+    "gate": "mtplx bench gemma-mtp --target-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-google-q8-g64-mlx --assistant-model /Users/youssof/Documents/MTPLX/models/gemma-4-31B-it-assistant-google-q8-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-targetq8-assistantq8-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-q8-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": 31.25938833784312,
+      "cache_memory_gib": 1.949219443835318,
+      "decode_s": 72.81026608298998,
+      "generated_tokens": 1000,
+      "mode": "ar",
+      "peak_memory_gib": 31.318915149196982,
+      "prefill_s": 2.473393041000236,
+      "tok_s": 13.73432695411645,
+      "token_preview": [
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761,
+        9996,
+        625,
+        24731,
+        236761
+      ]
+    },
+    "ar_confirmation": [
+      {
+        "active_memory_gib": 31.25938833784312,
+        "cache_memory_gib": 1.949219443835318,
+        "decode_s": 72.81026608298998,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 31.318915149196982,
+        "prefill_s": 2.473393041000236,
+        "tok_s": 13.73432695411645,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      },
+      {
+        "active_memory_gib": 31.749414331279695,
+        "cache_memory_gib": 2.0085750371217728,
+        "confirmation_repeat": 1,
+        "decode_s": 76.81720787500672,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 31.80895305145532,
+        "prefill_s": 0.23355866699421313,
+        "tok_s": 13.017916527598246,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      },
+      {
+        "active_memory_gib": 31.749414331279695,
+        "cache_memory_gib": 2.0085750371217728,
+        "confirmation_repeat": 2,
+        "decode_s": 85.77613120799651,
+        "generated_tokens": 1000,
+        "mode": "ar",
+        "peak_memory_gib": 31.80895305145532,
+        "prefill_s": 0.2485222500108648,
+        "tok_s": 11.658254877165346,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ]
+      }
+    ],
+    "best_block_confirmation": [
+      {
+        "acceptance": 0.9976047904191617,
+        "accepted_drafts": 833,
+        "active_memory_gib": 31.81573427375406,
+        "block_size": 6,
+        "cache_memory_gib": 66.49232691712677,
+        "decode_s": 29.219117750006262,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 835,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.937621283344924,
+        "prefill_s": 0.584632291996968,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 2.491870791785778,
+        "target_distribution_modes": {
+          "batched_logits": 167
+        },
+        "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": 0.009512208969681524,
+          "draft": 1.5696202969993465,
+          "rollback": 9.41639591474086e-05,
+          "target_distribution": 27.45037787995534,
+          "target_hidden": 0.0,
+          "verify": 0.17939917097100988
+        },
+        "tok_s": 34.22416818179891,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 167
+      },
+      {
+        "acceptance": 0.9976047904191617,
+        "accepted_drafts": 833,
+        "active_memory_gib": 31.81573427375406,
+        "block_size": 6,
+        "cache_memory_gib": 66.46690577454865,
+        "confirmation_repeat": 1,
+        "decode_s": 30.4154408339964,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 835,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.937621283344924,
+        "prefill_s": 0.26414108400058467,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 2.5255990302512874,
+        "target_distribution_modes": {
+          "batched_logits": 167
+        },
+        "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": 0.009377048874739558,
+          "draft": 1.6042620869993698,
+          "rollback": 8.99219885468483e-05,
+          "target_distribution": 28.613301041928935,
+          "target_hidden": 0.0,
+          "verify": 0.1785685370414285
+        },
+        "tok_s": 32.87803735799434,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 167
+      },
+      {
+        "acceptance": 0.9976047904191617,
+        "accepted_drafts": 833,
+        "active_memory_gib": 31.81573427375406,
+        "block_size": 6,
+        "cache_memory_gib": 66.46690577454865,
+        "confirmation_repeat": 2,
+        "decode_s": 30.19647025001177,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 835,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.937621283344924,
+        "prefill_s": 0.23273954199976288,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 2.840601252325612,
+        "target_distribution_modes": {
+          "batched_logits": 167
+        },
+        "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": 0.00978036891319789,
+          "draft": 1.6035626278899144,
+          "rollback": 9.504603804089129e-05,
+          "target_distribution": 28.391074630984804,
+          "target_hidden": 0.0,
+          "verify": 0.1814885419298662
+        },
+        "tok_s": 33.11645340400705,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 167
+      }
+    ],
+    "best_block_size": 6,
+    "best_speedup": 2.491870791785778,
+    "blockers": [],
+    "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": 7823,
+        "actual_min_rpm": 7234,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 7234,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 7234,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 7245
+            },
+            "rpm": 7234,
+            "target_rpm": 7245
+          },
+          {
+            "actual_rpm": 7823,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 7823,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 7824
+            },
+            "rpm": 7823,
+            "target_rpm": 7824
+          }
+        ],
+        "max_rpm": 7823,
+        "min_rpm": 7234,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7234,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 7245\n    },\n    {\n      \"actual_rpm\" : 7823,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 7824\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.4,\n    \"TB0T\" : 33,\n    \"TCDX\" : 67.7,\n    \"TCHP\" : 60.9,\n    \"TCMb\" : 75.5,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 40.4,\n    \"TMVR\" : 61.7,\n    \"TPDX\" : 67.1,\n    \"TRDX\" : 68.2,\n    \"TS0P\" : 69.2,\n    \"Tg0j\" : 65.9,\n    \"Tm08\" : 65,\n    \"Tp04\" : 65.4,\n    \"Tp08\" : 65.1,\n    \"Tp0C\" : 65.1,\n    \"Tp0G\" : 65.6,\n    \"Tp0X\" : 65.6\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\" : 7234,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 7245\n    },\n    {\n      \"actual_rpm\" : 7823,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 7824\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.4,\n    \"TB0T\" : 33,\n    \"TCDX\" : 67.7,\n    \"TCHP\" : 60.9,\n    \"TCMb\" : 75.5,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 40.4,\n    \"TMVR\" : 61.7,\n    \"TPDX\" : 67.1,\n    \"TRDX\" : 68.2,\n    \"TS0P\" : 69.2,\n    \"Tg0j\" : 65.9,\n    \"Tm08\" : 65,\n    \"Tp04\" : 65.4,\n    \"Tp08\" : 65.1,\n    \"Tp0C\" : 65.1,\n    \"Tp0G\" : 65.6,\n    \"Tp0X\" : 65.6\n  }\n}"
+          }
+        },
+        "target_max_rpm": 7824,
+        "target_min_rpm": 7245
+      },
+      "message": "fan restore was attempted but not verified",
+      "ok": false,
+      "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": 7424,
+        "actual_min_rpm": 7357,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 7357,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7357,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7357,
+            "target_rpm": 7826
+          },
+          {
+            "actual_rpm": 7424,
+            "max_capacity_rpm": 7826,
+            "mode": "manual",
+            "raw": {
+              "actual_rpm": 7424,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "manual",
+              "target_rpm": 7826
+            },
+            "rpm": 7424,
+            "target_rpm": 7826
+          }
+        ],
+        "max_rpm": 7424,
+        "min_rpm": 7357,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 7357,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7424,\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\" : 33,\n    \"TCDX\" : 66.4,\n    \"TCHP\" : 59.5,\n    \"TCMb\" : 73.7,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 39.1,\n    \"TMVR\" : 59,\n    \"TPDX\" : 64,\n    \"TRDX\" : 66.9,\n    \"TS0P\" : 66,\n    \"Tg0j\" : 64.4,\n    \"Tm08\" : 63.9,\n    \"Tp04\" : 64.4,\n    \"Tp08\" : 64.1,\n    \"Tp0C\" : 64.3,\n    \"Tp0G\" : 64.8,\n    \"Tp0X\" : 64.4\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\" : 7357,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"manual\",\n      \"target_rpm\" : 7826\n    },\n    {\n      \"actual_rpm\" : 7424,\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\" : 33,\n    \"TCDX\" : 66.4,\n    \"TCHP\" : 59.5,\n    \"TCMb\" : 73.7,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 39.1,\n    \"TMVR\" : 59,\n    \"TPDX\" : 64,\n    \"TRDX\" : 66.9,\n    \"TS0P\" : 66,\n    \"Tg0j\" : 64.4,\n    \"Tm08\" : 63.9,\n    \"Tp04\" : 64.4,\n    \"Tp08\" : 64.1,\n    \"Tp0C\" : 64.3,\n    \"Tp0G\" : 64.8,\n    \"Tp0X\" : 64.4\n  }\n}"
+          }
+        },
+        "target_max_rpm": 7826,
+        "target_min_rpm": 7826
+      },
+      "baseline": {
+        "actual_max_rpm": 6020,
+        "actual_min_rpm": 5580,
+        "capacity_max_rpm": 7826,
+        "capacity_min_rpm": 7826,
+        "fans": [
+          {
+            "actual_rpm": 5580,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 5580,
+              "index": 0,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 5575
+            },
+            "rpm": 5580,
+            "target_rpm": 5575
+          },
+          {
+            "actual_rpm": 6020,
+            "max_capacity_rpm": 7826,
+            "mode": "auto",
+            "raw": {
+              "actual_rpm": 6020,
+              "index": 1,
+              "max_rpm": 7826,
+              "min_rpm": 2317,
+              "mode": "auto",
+              "target_rpm": 6021
+            },
+            "rpm": 6020,
+            "target_rpm": 6021
+          }
+        ],
+        "max_rpm": 6020,
+        "min_rpm": 5580,
+        "ok": true,
+        "raw": {
+          "attempts": [
+            {
+              "command": [
+                "/Users/youssof/.mtplx/bin/thermalforge",
+                "status"
+              ],
+              "ok": true,
+              "returncode": 0,
+              "stderr": "",
+              "stdout": "{\n  \"fans\" : [\n    {\n      \"actual_rpm\" : 5580,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5575\n    },\n    {\n      \"actual_rpm\" : 6020,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6021\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.6,\n    \"TB0T\" : 33,\n    \"TCDX\" : 66.6,\n    \"TCHP\" : 59.7,\n    \"TCMb\" : 74.3,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 39.1,\n    \"TMVR\" : 59.5,\n    \"TPDX\" : 64.9,\n    \"TRDX\" : 67.6,\n    \"TS0P\" : 67,\n    \"Tg0j\" : 64.9,\n    \"Tm08\" : 64.5,\n    \"Tp04\" : 64.3,\n    \"Tp08\" : 64,\n    \"Tp0C\" : 63.9,\n    \"Tp0G\" : 64.7,\n    \"Tp0X\" : 64.5\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\" : 5580,\n      \"index\" : 0,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 5575\n    },\n    {\n      \"actual_rpm\" : 6020,\n      \"index\" : 1,\n      \"max_rpm\" : 7826,\n      \"min_rpm\" : 2317,\n      \"mode\" : \"auto\",\n      \"target_rpm\" : 6021\n    }\n  ],\n  \"temperatures\" : {\n    \"TAOL\" : 34.6,\n    \"TB0T\" : 33,\n    \"TCDX\" : 66.6,\n    \"TCHP\" : 59.7,\n    \"TCMb\" : 74.3,\n    \"TG0B\" : 33,\n    \"TG0H\" : 33,\n    \"TG0V\" : 33,\n    \"TH0x\" : 39.1,\n    \"TMVR\" : 59.5,\n    \"TPDX\" : 64.9,\n    \"TRDX\" : 67.6,\n    \"TS0P\" : 67,\n    \"Tg0j\" : 64.9,\n    \"Tm08\" : 64.5,\n    \"Tp04\" : 64.3,\n    \"Tp08\" : 64,\n    \"Tp0C\" : 63.9,\n    \"Tp0G\" : 64.7,\n    \"Tp0X\" : 64.5\n  }\n}"
+          }
+        },
+        "target_max_rpm": 6021,
+        "target_min_rpm": 5575
+      },
+      "message": "fans ramped to max (actual 7357-7424 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: 5580 RPM \u2192 max (7826 RPM)\nFan 1: 6020 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": 2.5439134852185834,
+    "memory_gate": {
+      "ar_peak_memory_gib": 31.80895305145532,
+      "lte_ar_plus_6_gib": true,
+      "lte_ar_times_1_18": true,
+      "mtp_peak_memory_gib": 31.937621283344924
+    },
+    "mtp_by_block_size": {
+      "3": {
+        "acceptance": 0.9970059880239521,
+        "accepted_drafts": 666,
+        "active_memory_gib": 31.813567525707185,
+        "block_size": 3,
+        "cache_memory_gib": 82.18594096973538,
+        "decode_s": 100.14749887499784,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 668,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.90066777355969,
+        "prefill_s": 0.4200237909972202,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 0.7270302993175131,
+        "target_distribution_modes": {
+          "batched_logits": 334
+        },
+        "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": 0.02364034196943976,
+          "draft": 2.0620414960430935,
+          "rollback": 0.00019853397679980844,
+          "target_distribution": 97.66836021310883,
+          "target_hidden": 0.0,
+          "verify": 0.3714816799911205
+        },
+        "tok_s": 9.985271836375869,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 334
+      },
+      "4": {
+        "acceptance": 1.0,
+        "accepted_drafts": 750,
+        "active_memory_gib": 31.815169698558748,
+        "block_size": 4,
+        "cache_memory_gib": 82.184341263026,
+        "decode_s": 88.162400583009,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 750,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.913927708752453,
+        "prefill_s": 0.45891812500485685,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 0.8258652849911423,
+        "target_distribution_modes": {
+          "batched_logits": 250
+        },
+        "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": 0.012636876970645972,
+          "draft": 2.101732614086359,
+          "rollback": 0.00013187690638005733,
+          "target_distribution": 85.75181395103573,
+          "target_hidden": 0.0,
+          "verify": 0.27962454402586445
+        },
+        "tok_s": 11.342703844122909,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 250
+      },
+      "5": {
+        "acceptance": 0.99875,
+        "accepted_drafts": 799,
+        "active_memory_gib": 31.814849263988435,
+        "block_size": 5,
+        "cache_memory_gib": 79.50701175443828,
+        "decode_s": 73.88629216598929,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 800,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.925461772829294,
+        "prefill_s": 0.5138492080004653,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 0.9854367291759348,
+        "target_distribution_modes": {
+          "batched_logits": 200
+        },
+        "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": 0.011033208094886504,
+          "draft": 1.9875367559725419,
+          "rollback": 0.0001222430873895064,
+          "target_distribution": 71.65346937690629,
+          "target_hidden": 0.0,
+          "verify": 0.22101375696365722
+        },
+        "tok_s": 13.534310231097393,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 200
+      },
+      "6": {
+        "acceptance": 0.9976047904191617,
+        "accepted_drafts": 833,
+        "active_memory_gib": 31.81573427375406,
+        "block_size": 6,
+        "cache_memory_gib": 66.49232691712677,
+        "decode_s": 29.219117750006262,
+        "draft_sampler": {
+          "temperature": 1.0,
+          "top_k": 64,
+          "top_p": 0.95
+        },
+        "drafted_tokens": 835,
+        "generated_tokens": 1000,
+        "mode": "mtp",
+        "peak_memory_gib": 31.937621283344924,
+        "prefill_s": 0.584632291996968,
+        "row_distribution_evals": 0,
+        "speedup_vs_ar": 2.491870791785778,
+        "target_distribution_modes": {
+          "batched_logits": 167
+        },
+        "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": 0.009512208969681524,
+          "draft": 1.5696202969993465,
+          "rollback": 9.41639591474086e-05,
+          "target_distribution": 27.45037787995534,
+          "target_hidden": 0.0,
+          "verify": 0.17939917097100988
+        },
+        "tok_s": 34.22416818179891,
+        "token_preview": [
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761,
+          9996,
+          625,
+          24731,
+          236761
+        ],
+        "verify_calls": 167
+      }
+    },
+    "passed": true,
+    "prompt_id": "flappy_html5_canvas_game",
+    "prompt_tokens": 119,
+    "sampler": {
+      "seed": 0,
+      "temperature": 1.0,
+      "top_k": 64,
+      "top_p": 0.95
+    }
+  },
+  "status": "passed"
+}

mtplx_pair.json

@@ -0,0 +1,53 @@
+{
+  "format_version": 1,
+  "name": "Gemma4 MTPLX Optimized Quality",
+  "variant": "quality",
+  "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": 8,
+      "group_size": 64,
+      "mode": "affine"
+    }
+  },
+  "assistant": {
+    "role": "drafter",
+    "model_type": "gemma4_assistant",
+    "quantization": {
+      "bits": 8,
+      "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": 833,
+      "drafted": 835,
+      "ratio": 0.9976047904191617
+    },
+    "observed_mtp_tok_s": [
+      34.22416818179891,
+      32.87803735799434,
+      33.11645340400705
+    ],
+    "speedup_vs_ar": 2.491870791785778
+  }
+}

target/README.md

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

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": 8,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 8,
+        "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-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3fe7c584ab9fb211c3071ac035628ccc6861c4a7e1b59c200f58b33ddb81435
+size 5258667341

target/model-00002-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed27cbf128295cb8c5fe462ccee517f5840f034b8310af432707c6b01089d2ce
+size 5328997408

target/model-00003-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a6cf32b03f6953bb42d863a4cc28e07289b6ea9fd501dfa9882b19e5539b048
+size 5270505988

target/model-00004-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:615bb5fb4e2150da2ef820f2beae5dcf7939fae8b6405c5d8cfb376a16d18ff3
+size 5329041180

target/model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1f2fad40a9535464bc81e522c13021d2478413be623cd07d2db00a2b19fc01d
+size 5346547174

target/model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dbf38950088a46ed5c9baab1921f5aa75ad4a498ee51e25765c951148983761
+size 5270505993

target/model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f0db64eef976c2985704d0aa7cbf4bc852d62a572bc59a02b20e62fe4f72806
+size 813126760

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>"
+}