Upload _reasoning_rl_multiseed/exp_phase8_reasoning_grounded_rl_t1_r128_alpha1_s3_20260506_234027/manifest.json with huggingface_hub
Browse files
_reasoning_rl_multiseed/exp_phase8_reasoning_grounded_rl_t1_r128_alpha1_s3_20260506_234027/manifest.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"trainer": "train_reasoning_grounded_rl.py",
|
| 3 |
+
"stage": "post_sft_rl",
|
| 4 |
+
"init_ckpt": "/workspace/dnathinker/runs/exp_phase5_stage_a_v7_reasoning_t1_r128_FIXED_20260502_104222/best.pt",
|
| 5 |
+
"llm_name": "Qwen/Qwen3.5-0.8B",
|
| 6 |
+
"lora_r": 128,
|
| 7 |
+
"lora_layers": [
|
| 8 |
+
3,
|
| 9 |
+
7,
|
| 10 |
+
11,
|
| 11 |
+
15,
|
| 12 |
+
19,
|
| 13 |
+
23
|
| 14 |
+
],
|
| 15 |
+
"lora_modules": [
|
| 16 |
+
"k_proj",
|
| 17 |
+
"o_proj",
|
| 18 |
+
"q_proj",
|
| 19 |
+
"v_proj"
|
| 20 |
+
],
|
| 21 |
+
"n_trainable": 8650752,
|
| 22 |
+
"alpha_halluc": 1.0,
|
| 23 |
+
"beta_tags": 0.1,
|
| 24 |
+
"gamma_consensus": 0.2,
|
| 25 |
+
"rollouts_per_prompt": 4,
|
| 26 |
+
"lr": 5e-06,
|
| 27 |
+
"num_rl_steps": 250,
|
| 28 |
+
"seed": 3
|
| 29 |
+
}
|