Flanker/slm-rl-space_invaders

pip install "transformers>=4.46" peft accelerate torch

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

BASE = "LiquidAI/LFM2.5-350M"
ADAPTER = "Flanker/slm-rl-space_invaders"  # 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 Space Invaders. 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))

from huggingface_hub import snapshot_download

path = snapshot_download("Flanker/slm-rl-space_invaders", allow_patterns="adapter/*")
# then: PeftModel.from_pretrained(base_model, f"{path}/adapter")

slm-rl evolve --game space-invaders \
  --dataset-url Flanker/slm-rl-space_invaders-data \
  --adapter-url Flanker/slm-rl-space_invaders \
  --generations 2

{
  "eval": {
    "episodes": 4,
    "intervention_rate": 0.0,
    "invalid_rate": 0.0,
    "mean_entropy": null,
    "mean_score": 0.7916666666666666,
    "primary": 0.7916666666666666,
    "win_rate": 0.0
  },
  "gate": {
    "promoted": true,
    "reason": "primary -1.9000 -> 0.7917, invalid_rate 0.0000, intervention_rate 0.0000"
  },
  "train": {
    "entropy": 0.607812587171793,
    "frac_reward_zero_std": 0.75,
    "kl": 0.08809181526885368,
    "loss": -0.004297492280602455,
    "num_prompts": 32,
    "reward": 0.09375
  }
}