Views
No views yet
rl_baseline| Parameter | Value |
|---|---|
model_id | qwen/Qwen3-4B |
dataset_path | results/data/leetcode_train_medhard_filtered_nohint.jsonl |
seed | 65 |
reward_funcs_kwargs | {"CorrectOrHintedCompileCode": {"allow_hint": false}} |
screening_funcs_kwargs | {} |
system_prompt | None |
learning_rate | 7e-05 |
beta | 0.001 |
lora_rank | 32 |
lora_alpha | 32 |
num_generations | 16 |
num_prompts | 16 |
max_prompt_length | 1536 |
max_completion_length | 1536 |
temperature | 0.7 |
top_p | 0.95 |
max_steps (checkpoint) | 200 |
1{
2 "run_id": "20251227_165255_leetcode_train_medhard_filtered_nohint_baseline",
3 "model_id": "qwen/Qwen3-4B",
4 "dataset_path": "results/data/leetcode_train_medhard_filtered_nohint.jsonl",
5 "eval_dataset_path": null,
6 "save_merged": false,
7 "extra_metadata": null,
8 "skip_save": false,
9 "resume_from_checkpoint": false,
10 "seed": 65,
11 "logging_steps": 1,
12 "report_to": "wandb",
13 "eval_strategy": "steps",
14 "save_strategy": "steps",
15 "save_only_model": true,
16 "save_total_limit": null,
17 "save_steps": 50,
18 "load_in_4bit": false,
19 "load_in_8bit": false,
20 "lora_rank": 32,
21 "lora_alpha": 32,
22 "lora_dropout": 0.0,
23 "lora_bias": "none",
24 "use_rslora": false,
25 "loftq_config": null,
26 "system_prompt": null,
27 "system_prompt_method": "replace",
28 "reward_funcs_kwargs": {
29 "CorrectOrHintedCompileCode": {
30 "allow_hint": false
31 }
32 },
33 "screening_funcs_kwargs": {},
34 "beta": 0.001,
35 "optim": "adamw_8bit",
36 "learning_rate": 7e-05,
37 "lr_scheduler_type": "cosine",
38 "warmup_ratio": null,
39 "warmup_steps": 10,
40 "weight_decay": 0.1,
41 "adam_beta1": 0.9,
42 "adam_beta2": 0.99,
43 "max_grad_norm": 1.0,
44 "num_train_epochs": 1,
45 "max_steps": 200,
46 "max_prompt_length": 1536,
47 "max_completion_length": 1536,
48 "dataloader_num_workers": 4,
49 "num_generations": 16,
50 "num_prompts": 16,
51 "per_device_batch_size": 8,
52 "auto_find_batch_size": true,
53 "enable_gradient_checkpointing": true,
54 "gpu_memory_utilization": 0.6,
55 "use_vllm": true,
56 "temperature": 0.7,
57 "top_p": 0.95,
58 "repetition_penalty": 1.0,
59 "generation_kwargs": {},
60 "enable_thinking": false,
61 "cache_activations": false,
62 "cache_activations_layers": [
63 18
64 ],
65 "cache_activations_position": "response_avg",
66 "fill_nan_global": true,
67 "log_completions": true,
68 "dataloader_prefetch_factor": 2,
69 "dataloader_persistent_workers": true,
70 "dataloader_pin_memory": true,
71 "max_steps (checkpoint)": 200
72}1from peft import PeftModel
2from transformers import AutoModelForCausalLM
3
4base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B")
5model = PeftModel.from_pretrained(base_model, "ariahw/rl-rewardhacking-leetcode-rl-baseline-s65")