Initial upload: GPT-2 compressed with BigSmall (lossless, 75.53% ratio)

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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
+model.bs filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,50 @@
+---
+license: mit
+tags:
+- bigsmall
+- compression
+- lossless
+- gpt2
+---
+
+# GPT-2 (BigSmall compressed)
+
+This is GPT-2 compressed with [BigSmall](https://github.com/wpferrell/Bigsmall) — lossless neural network weight compression.
+
+The weights are **bit-for-bit identical** to the original after decompression. No quality degradation, no accuracy loss.
+
+## Usage
+
+```python
+pip install bigsmall
+```
+
+```python
+import bigsmall
+from transformers import GPT2LMHeadModel, GPT2Tokenizer
+
+# Load compressed weights
+state_dict = bigsmall.from_pretrained("wpferrell/gpt2-bigsmall")
+
+# Load into model
+model = GPT2LMHeadModel.from_pretrained("gpt2", state_dict=state_dict)
+tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
+
+# Run inference - identical to original GPT-2
+inputs = tokenizer("Hello, I'm a language model", return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=20)
+print(tokenizer.decode(outputs[0]))
+```
+
+## Compression stats
+- Original size: ~548 MB
+- Compressed size: ~414 MB  
+- Ratio: 75.53% (lossless)
+- Format: FP32
+- Round-trip verified: 160/160 tensors md5-identical
+
+## About BigSmall
+BigSmall compresses model weights losslessly. Unlike quantization, the decompressed weights are bit-for-bit identical to the originals. Supports BF16, FP16, FP32, FP64, INT8 formats across LLMs and diffusion models.
+
+- PyPI: `pip install bigsmall`
+- GitHub: https://github.com/wpferrell/Bigsmall

bigsmall.index.json

@@ -0,0 +1,179 @@
+{
+  "metadata": {
+    "bigsmall_version": "1.0.0",
+    "container_version": 1,
+    "format": "fp32",
+    "mode": "balanced",
+    "model_type": "llm",
+    "total_size": 413973499,
+    "total_raw_size": 548090880,
+    "ratio_pct": 75.53008344163655,
+    "shard_count": 1,
+    "tensor_count": 160,
+    "shards": [
+      "model.bs"
+    ]
+  },
+  "weight_map": {
+    "h.0.attn.bias": "model.bs",
+    "h.0.attn.c_attn.bias": "model.bs",
+    "h.0.attn.c_attn.weight": "model.bs",
+    "h.0.attn.c_proj.bias": "model.bs",
+    "h.0.attn.c_proj.weight": "model.bs",
+    "h.0.ln_1.bias": "model.bs",
+    "h.0.ln_1.weight": "model.bs",
+    "h.0.ln_2.bias": "model.bs",
+    "h.0.ln_2.weight": "model.bs",
+    "h.0.mlp.c_fc.bias": "model.bs",
+    "h.0.mlp.c_fc.weight": "model.bs",
+    "h.0.mlp.c_proj.bias": "model.bs",
+    "h.0.mlp.c_proj.weight": "model.bs",
+    "h.1.attn.bias": "model.bs",
+    "h.1.attn.c_attn.bias": "model.bs",
+    "h.1.attn.c_attn.weight": "model.bs",
+    "h.1.attn.c_proj.bias": "model.bs",
+    "h.1.attn.c_proj.weight": "model.bs",
+    "h.1.ln_1.bias": "model.bs",
+    "h.1.ln_1.weight": "model.bs",
+    "h.1.ln_2.bias": "model.bs",
+    "h.1.ln_2.weight": "model.bs",
+    "h.1.mlp.c_fc.bias": "model.bs",
+    "h.1.mlp.c_fc.weight": "model.bs",
+    "h.1.mlp.c_proj.bias": "model.bs",
+    "h.1.mlp.c_proj.weight": "model.bs",
+    "h.10.attn.bias": "model.bs",
+    "h.10.attn.c_attn.bias": "model.bs",
+    "h.10.attn.c_attn.weight": "model.bs",
+    "h.10.attn.c_proj.bias": "model.bs",
+    "h.10.attn.c_proj.weight": "model.bs",
+    "h.10.ln_1.bias": "model.bs",
+    "h.10.ln_1.weight": "model.bs",
+    "h.10.ln_2.bias": "model.bs",
+    "h.10.ln_2.weight": "model.bs",
+    "h.10.mlp.c_fc.bias": "model.bs",
+    "h.10.mlp.c_fc.weight": "model.bs",
+    "h.10.mlp.c_proj.bias": "model.bs",
+    "h.10.mlp.c_proj.weight": "model.bs",
+    "h.11.attn.bias": "model.bs",
+    "h.11.attn.c_attn.bias": "model.bs",
+    "h.11.attn.c_attn.weight": "model.bs",
+    "h.11.attn.c_proj.bias": "model.bs",
+    "h.11.attn.c_proj.weight": "model.bs",
+    "h.11.ln_1.bias": "model.bs",
+    "h.11.ln_1.weight": "model.bs",
+    "h.11.ln_2.bias": "model.bs",
+    "h.11.ln_2.weight": "model.bs",
+    "h.11.mlp.c_fc.bias": "model.bs",
+    "h.11.mlp.c_fc.weight": "model.bs",
+    "h.11.mlp.c_proj.bias": "model.bs",
+    "h.11.mlp.c_proj.weight": "model.bs",
+    "h.2.attn.bias": "model.bs",
+    "h.2.attn.c_attn.bias": "model.bs",
+    "h.2.attn.c_attn.weight": "model.bs",
+    "h.2.attn.c_proj.bias": "model.bs",
+    "h.2.attn.c_proj.weight": "model.bs",
+    "h.2.ln_1.bias": "model.bs",
+    "h.2.ln_1.weight": "model.bs",
+    "h.2.ln_2.bias": "model.bs",
+    "h.2.ln_2.weight": "model.bs",
+    "h.2.mlp.c_fc.bias": "model.bs",
+    "h.2.mlp.c_fc.weight": "model.bs",
+    "h.2.mlp.c_proj.bias": "model.bs",
+    "h.2.mlp.c_proj.weight": "model.bs",
+    "h.3.attn.bias": "model.bs",
+    "h.3.attn.c_attn.bias": "model.bs",
+    "h.3.attn.c_attn.weight": "model.bs",
+    "h.3.attn.c_proj.bias": "model.bs",
+    "h.3.attn.c_proj.weight": "model.bs",
+    "h.3.ln_1.bias": "model.bs",
+    "h.3.ln_1.weight": "model.bs",
+    "h.3.ln_2.bias": "model.bs",
+    "h.3.ln_2.weight": "model.bs",
+    "h.3.mlp.c_fc.bias": "model.bs",
+    "h.3.mlp.c_fc.weight": "model.bs",
+    "h.3.mlp.c_proj.bias": "model.bs",
+    "h.3.mlp.c_proj.weight": "model.bs",
+    "h.4.attn.bias": "model.bs",
+    "h.4.attn.c_attn.bias": "model.bs",
+    "h.4.attn.c_attn.weight": "model.bs",
+    "h.4.attn.c_proj.bias": "model.bs",
+    "h.4.attn.c_proj.weight": "model.bs",
+    "h.4.ln_1.bias": "model.bs",
+    "h.4.ln_1.weight": "model.bs",
+    "h.4.ln_2.bias": "model.bs",
+    "h.4.ln_2.weight": "model.bs",
+    "h.4.mlp.c_fc.bias": "model.bs",
+    "h.4.mlp.c_fc.weight": "model.bs",
+    "h.4.mlp.c_proj.bias": "model.bs",
+    "h.4.mlp.c_proj.weight": "model.bs",
+    "h.5.attn.bias": "model.bs",
+    "h.5.attn.c_attn.bias": "model.bs",
+    "h.5.attn.c_attn.weight": "model.bs",
+    "h.5.attn.c_proj.bias": "model.bs",
+    "h.5.attn.c_proj.weight": "model.bs",
+    "h.5.ln_1.bias": "model.bs",
+    "h.5.ln_1.weight": "model.bs",
+    "h.5.ln_2.bias": "model.bs",
+    "h.5.ln_2.weight": "model.bs",
+    "h.5.mlp.c_fc.bias": "model.bs",
+    "h.5.mlp.c_fc.weight": "model.bs",
+    "h.5.mlp.c_proj.bias": "model.bs",
+    "h.5.mlp.c_proj.weight": "model.bs",
+    "h.6.attn.bias": "model.bs",
+    "h.6.attn.c_attn.bias": "model.bs",
+    "h.6.attn.c_attn.weight": "model.bs",
+    "h.6.attn.c_proj.bias": "model.bs",
+    "h.6.attn.c_proj.weight": "model.bs",
+    "h.6.ln_1.bias": "model.bs",
+    "h.6.ln_1.weight": "model.bs",
+    "h.6.ln_2.bias": "model.bs",
+    "h.6.ln_2.weight": "model.bs",
+    "h.6.mlp.c_fc.bias": "model.bs",
+    "h.6.mlp.c_fc.weight": "model.bs",
+    "h.6.mlp.c_proj.bias": "model.bs",
+    "h.6.mlp.c_proj.weight": "model.bs",
+    "h.7.attn.bias": "model.bs",
+    "h.7.attn.c_attn.bias": "model.bs",
+    "h.7.attn.c_attn.weight": "model.bs",
+    "h.7.attn.c_proj.bias": "model.bs",
+    "h.7.attn.c_proj.weight": "model.bs",
+    "h.7.ln_1.bias": "model.bs",
+    "h.7.ln_1.weight": "model.bs",
+    "h.7.ln_2.bias": "model.bs",
+    "h.7.ln_2.weight": "model.bs",
+    "h.7.mlp.c_fc.bias": "model.bs",
+    "h.7.mlp.c_fc.weight": "model.bs",
+    "h.7.mlp.c_proj.bias": "model.bs",
+    "h.7.mlp.c_proj.weight": "model.bs",
+    "h.8.attn.bias": "model.bs",
+    "h.8.attn.c_attn.bias": "model.bs",
+    "h.8.attn.c_attn.weight": "model.bs",
+    "h.8.attn.c_proj.bias": "model.bs",
+    "h.8.attn.c_proj.weight": "model.bs",
+    "h.8.ln_1.bias": "model.bs",
+    "h.8.ln_1.weight": "model.bs",
+    "h.8.ln_2.bias": "model.bs",
+    "h.8.ln_2.weight": "model.bs",
+    "h.8.mlp.c_fc.bias": "model.bs",
+    "h.8.mlp.c_fc.weight": "model.bs",
+    "h.8.mlp.c_proj.bias": "model.bs",
+    "h.8.mlp.c_proj.weight": "model.bs",
+    "h.9.attn.bias": "model.bs",
+    "h.9.attn.c_attn.bias": "model.bs",
+    "h.9.attn.c_attn.weight": "model.bs",
+    "h.9.attn.c_proj.bias": "model.bs",
+    "h.9.attn.c_proj.weight": "model.bs",
+    "h.9.ln_1.bias": "model.bs",
+    "h.9.ln_1.weight": "model.bs",
+    "h.9.ln_2.bias": "model.bs",
+    "h.9.ln_2.weight": "model.bs",
+    "h.9.mlp.c_fc.bias": "model.bs",
+    "h.9.mlp.c_fc.weight": "model.bs",
+    "h.9.mlp.c_proj.bias": "model.bs",
+    "h.9.mlp.c_proj.weight": "model.bs",
+    "ln_f.bias": "model.bs",
+    "ln_f.weight": "model.bs",
+    "wpe.weight": "model.bs",
+    "wte.weight": "model.bs"
+  }
+}

config.json

@@ -0,0 +1,31 @@
+{
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50256,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 768,
+  "n_head": 12,
+  "n_layer": 12,
+  "n_positions": 1024,
+  "resid_pdrop": 0.1,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "text-generation": {
+      "do_sample": true,
+      "max_length": 50
+    }
+  },
+  "vocab_size": 50257
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 50256,
+  "eos_token_id": 50256,
+  "transformers_version": "4.26.0.dev0",
+  "_from_model_config": true
+}

model.bs

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1a8e938d66a1b5e046ca50d5334684c2a86aa9948f5f32a1bbe4743df5c89dd
+size 413973499

tokenizer_config.json

@@ -0,0 +1 @@
+{"model_max_length": 1024}