boxing: proper model card + transformers/PEFT usage

README.md

@@ -1,11 +1,115 @@
-# slm-rl-boxing
+---
+library_name: peft
+base_model: LiquidAI/LFM2.5-350M
+pipeline_tag: text-generation
+tags:
+  - lora
+  - peft
+  - transformers
+  - reinforcement-learning
+  - atari
+  - slm-rl
+  - boxing
+license: apache-2.0
+---
+# BLANK/slm-rl-boxing
 
-SFT LoRA for ALE Boxing (SLM-RL workshop warm-start).
+PEFT LoRA adapter that warm-starts **Boxing** play for
+[LiquidAI/LFM2.5-350M](https://huggingface.co/LiquidAI/LFM2.5-350M)
+in the [SLM-RL](https://github.com/CraftsMan-Labs/SLM-RL) workshop.
 
-- **Game:** boxing
-- **Base model:** LiquidAI/LFM2.5-350M
-- **Dataset:** BLANK/slm-rl-boxing
-- **Training:** reject_sft on DQN teacher demos
-- **Layout:**  — paste this repo as the playground adapter URL
+| | |
+|---|---|
+| **Game** | `boxing` |
+| **Base model** | `LiquidAI/LFM2.5-350M` |
+| **Adapter layout** | `adapter/` (PEFT `adapter_config.json` + weights) |
+| **Training** | `reject_sft` on DQN teacher demos |
+| **Champion generation** | 1 |
+| **Promoted** | True (reject_sft warm-start) |
+| **Dataset pack** | [BLANK/slm-rl-boxing](https://huggingface.co/datasets/BLANK/slm-rl-boxing) |
+| **DQN teacher** | [BLANK/slm-rl-boxing-dqn](https://huggingface.co/BLANK/slm-rl-boxing-dqn) |
+
+Paste `BLANK/slm-rl-boxing` as the playground **adapter URL** (and usually the same id
+as the **dataset URL**).
+
+## Install
+
+```bash
+pip install "transformers>=4.46" peft accelerate torch
+```
+
+## Load with transformers + PEFT
+
+Weights live under the `adapter/` subfolder — pass `subfolder="adapter"`.
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+BASE = "LiquidAI/LFM2.5-350M"
+ADAPTER = "BLANK/slm-rl-boxing"  # this repo
+
+device = (
+    "cuda" if torch.cuda.is_available()
+    else "mps" if torch.backends.mps.is_available()
+    else "cpu"
+)
+dtype = torch.bfloat16 if device != "cpu" else torch.float32
+
+tokenizer = AutoTokenizer.from_pretrained(BASE)
+if tokenizer.pad_token is None:
+    tokenizer.pad_token = tokenizer.eos_token
+
+model = AutoModelForCausalLM.from_pretrained(BASE, torch_dtype=dtype)
+model = PeftModel.from_pretrained(model, ADAPTER, subfolder="adapter")
+model.to(device).eval()
+
+messages = [
+    {"role": "system", "content": "You play Boxing. Reply with ACTION: <id>."},
+    {"role": "user", "content": "Legal actions: 1) NOOP 2) UP\nChoose."},
+]
+prompt = tokenizer.apply_chat_template(
+    messages, add_generation_prompt=True, tokenize=False,
+)
+inputs = tokenizer(prompt, return_tensors="pt").to(device)
+with torch.inference_mode():
+    out = model.generate(**inputs, max_new_tokens=24, do_sample=False)
+print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
+```
+
+### Download only the adapter files
+
+```python
+from huggingface_hub import snapshot_download
+
+path = snapshot_download("BLANK/slm-rl-boxing", allow_patterns="adapter/*")
+# then: PeftModel.from_pretrained(base_model, f"{path}/adapter")
+```
+
+## Workshop / SLM-RL CLI
+
+```bash
+slm-rl evolve --game boxing \
+  --dataset-url BLANK/slm-rl-boxing \
+  --adapter-url BLANK/slm-rl-boxing \
+  --dqn-url BLANK/slm-rl-boxing-dqn \
+  --generations 2
+```
+
+## Train metrics (if recorded)
+
+```json
+{
+  "eval": {
+    "skipped": true
+  },
+  "gate": {
+    "promoted": true,
+    "reason": "reject_sft warm-start"
+  },
+  "train": {}
+}
+```
 
 Trained with [SLM-RL](https://github.com/CraftsMan-Labs/SLM-RL).