RL-MPQ Balanced — 2026-06-11T20:29:45.309427

.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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,87 @@
+---
+license: gemma
+base_model: google/gemma-2-9b
+pipeline_tag: text-generation
+language:
+  - en
+tags:
+  - gemma
+  - text-generation
+  - rl-mpq
+  - mixed-precision
+  - quantization
+  - fake-quantization
+  - balanced
+library_name: transformers
+datasets:
+  - wikitext
+widget:
+  - text: "The capital of France is"
+---
+
+# Gemma 2 9B — RL-MPQ Balanced
+
+Standalone **RL-MPQ** (Reinforcement Learning Mixed-Precision Quantization) checkpoint for the
+**Balanced** scenario — a quantized variant of
+[google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b).
+
+| Field | Value |
+|-------|-------|
+| **Base model** | [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b) |
+| **Scenario** | Balanced |
+| **Avg bits / weight** | 4.2857 |
+| **Compression vs FP16** | 3.7333× |
+| **WikiText-2 PPL** | 127.0798 |
+| **Layers** | 42 |
+| **Bit distribution** | `{'4': 39, '8': 3}` |
+| **Format** | Fake-quant FP16 + `rlmpq_policy.json` |
+
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+repo = "AvoCahDoe/gemma-2-9b-rlmpq-balanced"
+
+model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="float16")
+tokenizer = AutoTokenizer.from_pretrained(repo)
+```
+
+## Other Gemma 2 9B scenarios
+
+| Scenario | Avg bits | Compression | WikiText-2 PPL |
+|----------|----------|-------------|----------------|
+| [Aggressive](https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq-aggressive) | 3.6667 | 4.3636x | 162.8437 |
+| [Conservative](https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq-conservative) | 5.1429 | 3.1111x | 116.5244 |
+| [Extreme Survival](https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq-extreme-survival) | 2.7857 | 5.7436x | 424.7991 |
+| [High Fidelity](https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq-high-fidelity) | 7.0476 | 2.2703x | 104.8098 |
+
+Grouped archive (all scenarios in one repo):
+[AvoCahDoe/gemma-2-9b-rlmpq](https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq)
+
+## Method
+
+1. **Phase 3** — PPO agent assigns per-layer bit widths under the Balanced reward target.
+2. **Phase 4** — Policy replayed on real weights; WikiText-2 perplexity validates quality.
+3. **Export** — Fake-quantized FP16 weights compatible with Hugging Face Transformers.
+
+## Files
+
+| File | Description |
+|------|-------------|
+| `config.json` | Llama architecture + RL-MPQ metadata |
+| `model.safetensors` | Fake-quantized weights |
+| `rlmpq_policy.json` | Per-layer bit-width policy |
+| `rlmpq_metrics.json` | Validation & PPL summary |
+
+## Citation
+
+```bibtex
+@misc{rlmpq_gemma_2_9b_balanced_2026,
+  title  = {RL-MPQ Balanced: Gemma 2 9B Mixed-Precision Quantization},
+  author = {AvoCahDoe},
+  year   = {2026},
+  url    = {https://huggingface.co/AvoCahDoe/gemma-2-9b-rlmpq-balanced}
+}
+```

config.json

@@ -0,0 +1,99 @@
+{
+  "architectures": [
+    "Gemma2ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": 50.0,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "dtype": "bfloat16",
+  "eos_token_id": 1,
+  "final_logit_softcapping": 30.0,
+  "head_dim": 256,
+  "hidden_act": "gelu_pytorch_tanh",
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 3584,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "layer_types": [
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 8192,
+  "model_type": "gemma2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 42,
+  "num_key_value_heads": 8,
+  "pad_token_id": 0,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 10000.0,
+    "rope_type": "default",
+    "type": "default"
+  },
+  "sliding_window": 4096,
+  "sliding_window_size": 4096,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.11.0",
+  "use_bidirectional_attention": null,
+  "use_cache": true,
+  "vocab_size": 256000,
+  "base_model": "google/gemma-2-9b",
+  "quantization": {
+    "method": "rl-mpq",
+    "format": "fake-quant-fp16",
+    "scenario": "Balanced",
+    "avg_bits": 4.2857,
+    "description": "Per-layer mixed bit-width via PPO-trained policy (not GPTQ/AWQ)"
+  },
+  "rlmpq": {
+    "framework": "RL-NMP-Model-Quantasation",
+    "scenario": "Balanced",
+    "scenario_label": "Balanced",
+    "repo_id": "AvoCahDoe/gemma-2-9b-rlmpq-balanced",
+    "grouped_repo": "AvoCahDoe/gemma-2-9b-rlmpq",
+    "collection_title": "RL-MPQ \u2014 Gemma 2 9B"
+  }
+}

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 1,
+  "pad_token_id": 0,
+  "transformers_version": "5.11.0"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76d0b084b91649e0ae30aa79e829ff7cec6901d8f9c7b15ae33bc9a226201788
+size 18483467000

rlmpq_metrics.json

@@ -0,0 +1,26 @@
+{
+  "framework": "RL-NMP-Model-Quantasation",
+  "method": "RL-MPQ",
+  "repo_id": "AvoCahDoe/gemma-2-9b-rlmpq-balanced",
+  "base_model": "google/gemma-2-9b",
+  "model_slug": "gemma_2_9b",
+  "scenario": "Balanced",
+  "scenario_label": "Balanced",
+  "avg_bits_per_weight": 4.2857,
+  "compression_vs_fp16": 3.7333,
+  "wikitext2_perplexity": 127.0798,
+  "bit_distribution": {
+    "4": 39,
+    "8": 3
+  },
+  "quantization": {
+    "type": "fake-quant-fp16",
+    "group_size": 128,
+    "scheme": "per-layer asymmetric group-wise",
+    "packed_format": false
+  },
+  "exported_at": "2026-06-11T20:29:44.975370",
+  "policy_source": "/workspace/RL-NMP-Model-Quantasation/phase3/models/gemma_2_9b/results/Balanced_policy.json",
+  "policy_path": "/workspace/RL-NMP-Model-Quantasation/phase3/models/gemma_2_9b/results/Balanced_policy.json",
+  "training_run": "20260610_202902"
+}

rlmpq_policy.json

@@ -0,0 +1,73 @@
+{
+  "slug": "gemma_2_9b",
+  "hf_id": "google/gemma-2-9b",
+  "scenario": "Balanced",
+  "params": {
+    "lambda_mse": 1.0,
+    "lambda_bit": 0.25,
+    "max_mse_clip": 1.0,
+    "target_bits": 4.5,
+    "lambda_budget": 5.0
+  },
+  "policy": [
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    4,
+    8,
+    8,
+    8
+  ],
+  "num_layers": 42,
+  "avg_bits": 4.2857,
+  "total_reward": -6.244206,
+  "compression_ratio": 3.7333,
+  "approx_size_gb": 4.821,
+  "bit_distribution": {
+    "4": 39,
+    "8": 3
+  },
+  "avg_mse_per_bit": {
+    "4": 0.02916564,
+    "8": 4.197e-05
+  },
+  "validation_elapsed_s": 0.034,
+  "policy_source": "/workspace/RL-NMP-Model-Quantasation/phase3/models/gemma_2_9b/results/Balanced_policy.json",
+  "policy_path": "/workspace/RL-NMP-Model-Quantasation/phase3/models/gemma_2_9b/results/Balanced_policy.json",
+  "training_run": "20260610_202902"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:487cee8724215dcd2dde8888539e8b1bf844ceb5dbbe27f7845abda69eeb060f
+size 34362872

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<eos>",
+  "extra_special_tokens": [
+    "<start_of_turn>",
+    "<end_of_turn>"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}