add rl_grpo/coin_real_4x_direct (copied from sidbaines/scimt-prior-coins-dispatch-sdf-aft-v1)
Browse files- .gitattributes +1 -0
- rl_grpo/coin_real_4x_direct/RL_TRAINED.json +13 -0
- rl_grpo/coin_real_4x_direct/checkpoint-128/adapter_config.json +191 -0
- rl_grpo/coin_real_4x_direct/checkpoint-128/adapter_model.safetensors +3 -0
- rl_grpo/coin_real_4x_direct/checkpoint-16/adapter_config.json +191 -0
- rl_grpo/coin_real_4x_direct/checkpoint-16/adapter_model.safetensors +3 -0
- rl_grpo/coin_real_4x_direct/checkpoint-256/adapter_config.json +191 -0
- rl_grpo/coin_real_4x_direct/checkpoint-256/adapter_model.safetensors +3 -0
- rl_grpo/coin_real_4x_direct/checkpoint-256/trainer_state.json +834 -0
- rl_grpo/coin_real_4x_direct/checkpoint-32/adapter_config.json +191 -0
- rl_grpo/coin_real_4x_direct/checkpoint-32/adapter_model.safetensors +3 -0
- rl_grpo/coin_real_4x_direct/checkpoint-64/adapter_config.json +191 -0
- rl_grpo/coin_real_4x_direct/checkpoint-64/adapter_model.safetensors +3 -0
- rl_grpo/coin_real_4x_direct/checkpoint.json +132 -0
- rl_grpo/coin_real_4x_direct/logs/raw_rollouts.rank-0.jsonl +3 -0
.gitattributes
CHANGED
|
@@ -135,3 +135,4 @@ sdf/4x/shared/post_dolci90/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
|
| 135 |
sdf/1x/shared/post_dolci90/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 136 |
gate2_midtrain4/balanced/post_dolci100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 137 |
gate2_midtrain4/dolmino/post_dolci100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 135 |
sdf/1x/shared/post_dolci90/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 136 |
gate2_midtrain4/balanced/post_dolci100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 137 |
gate2_midtrain4/dolmino/post_dolci100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 138 |
+
rl_grpo/coin_real_4x_direct/logs/raw_rollouts.rank-0.jsonl filter=lfs diff=lfs merge=lfs -text
|
rl_grpo/coin_real_4x_direct/RL_TRAINED.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "dispatch_rl_v3",
|
| 3 |
+
"label": "coin_real_4x_direct",
|
| 4 |
+
"mode": "direct",
|
| 5 |
+
"parent": "/workspace/rl3_direct_b/parent",
|
| 6 |
+
"dataset": "/workspace/rl3_direct_b/data/direct/train.jsonl",
|
| 7 |
+
"episodes": 8192,
|
| 8 |
+
"adapter": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/sampler",
|
| 9 |
+
"minutes": 49.87,
|
| 10 |
+
"dropped_overlong": 0,
|
| 11 |
+
"temperature": 0.7,
|
| 12 |
+
"learning_rate": 1e-05
|
| 13 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-128/adapter_config.json
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/workspace/rl3_direct_b/parent",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"layer_replication": null,
|
| 16 |
+
"layers_pattern": null,
|
| 17 |
+
"layers_to_transform": null,
|
| 18 |
+
"loftq_config": {},
|
| 19 |
+
"lora_alpha": 64,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 32,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"language_model.layers.13.self_attn.q_proj",
|
| 35 |
+
"46.self_attn.q_proj",
|
| 36 |
+
"39.self_attn.v_proj",
|
| 37 |
+
"language_model.layers.15.self_attn.q_proj",
|
| 38 |
+
"language_model.layers.19.self_attn.k_proj",
|
| 39 |
+
"language_model.layers.21.self_attn.v_proj",
|
| 40 |
+
"language_model.layers.9.self_attn.q_proj",
|
| 41 |
+
"41.self_attn.q_proj",
|
| 42 |
+
"o_proj",
|
| 43 |
+
"language_model.layers.16.self_attn.q_proj",
|
| 44 |
+
"language_model.layers.13.self_attn.k_proj",
|
| 45 |
+
"40.self_attn.k_proj",
|
| 46 |
+
"42.self_attn.k_proj",
|
| 47 |
+
"language_model.layers.18.self_attn.q_proj",
|
| 48 |
+
"language_model.layers.12.self_attn.k_proj",
|
| 49 |
+
"34.self_attn.k_proj",
|
| 50 |
+
"language_model.layers.0.self_attn.q_proj",
|
| 51 |
+
"34.self_attn.q_proj",
|
| 52 |
+
"37.self_attn.v_proj",
|
| 53 |
+
"language_model.layers.12.self_attn.q_proj",
|
| 54 |
+
"33.self_attn.k_proj",
|
| 55 |
+
"language_model.layers.8.self_attn.v_proj",
|
| 56 |
+
"32.self_attn.q_proj",
|
| 57 |
+
"43.self_attn.v_proj",
|
| 58 |
+
"language_model.layers.6.self_attn.q_proj",
|
| 59 |
+
"language_model.layers.7.self_attn.k_proj",
|
| 60 |
+
"language_model.layers.19.self_attn.q_proj",
|
| 61 |
+
"32.self_attn.k_proj",
|
| 62 |
+
"language_model.layers.23.self_attn.q_proj",
|
| 63 |
+
"language_model.layers.3.self_attn.q_proj",
|
| 64 |
+
"language_model.layers.22.self_attn.v_proj",
|
| 65 |
+
"36.self_attn.q_proj",
|
| 66 |
+
"33.self_attn.v_proj",
|
| 67 |
+
"language_model.layers.16.self_attn.k_proj",
|
| 68 |
+
"language_model.layers.1.self_attn.k_proj",
|
| 69 |
+
"language_model.layers.8.self_attn.k_proj",
|
| 70 |
+
"language_model.layers.15.self_attn.k_proj",
|
| 71 |
+
"33.self_attn.q_proj",
|
| 72 |
+
"language_model.layers.15.self_attn.v_proj",
|
| 73 |
+
"language_model.layers.26.self_attn.v_proj",
|
| 74 |
+
"42.self_attn.v_proj",
|
| 75 |
+
"29.self_attn.q_proj",
|
| 76 |
+
"45.self_attn.v_proj",
|
| 77 |
+
"language_model.layers.17.self_attn.q_proj",
|
| 78 |
+
"language_model.layers.24.self_attn.k_proj",
|
| 79 |
+
"language_model.layers.1.self_attn.v_proj",
|
| 80 |
+
"language_model.layers.20.self_attn.k_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"language_model.layers.5.self_attn.q_proj",
|
| 83 |
+
"34.self_attn.v_proj",
|
| 84 |
+
"44.self_attn.v_proj",
|
| 85 |
+
"language_model.layers.0.self_attn.k_proj",
|
| 86 |
+
"28.self_attn.v_proj",
|
| 87 |
+
"language_model.layers.22.self_attn.q_proj",
|
| 88 |
+
"30.self_attn.k_proj",
|
| 89 |
+
"30.self_attn.q_proj",
|
| 90 |
+
"language_model.layers.6.self_attn.v_proj",
|
| 91 |
+
"language_model.layers.14.self_attn.k_proj",
|
| 92 |
+
"language_model.layers.5.self_attn.v_proj",
|
| 93 |
+
"41.self_attn.v_proj",
|
| 94 |
+
"language_model.layers.25.self_attn.k_proj",
|
| 95 |
+
"31.self_attn.k_proj",
|
| 96 |
+
"language_model.layers.5.self_attn.k_proj",
|
| 97 |
+
"language_model.layers.4.self_attn.k_proj",
|
| 98 |
+
"35.self_attn.k_proj",
|
| 99 |
+
"37.self_attn.k_proj",
|
| 100 |
+
"language_model.layers.11.self_attn.k_proj",
|
| 101 |
+
"language_model.layers.8.self_attn.q_proj",
|
| 102 |
+
"29.self_attn.k_proj",
|
| 103 |
+
"47.self_attn.q_proj",
|
| 104 |
+
"language_model.layers.13.self_attn.v_proj",
|
| 105 |
+
"language_model.layers.9.self_attn.v_proj",
|
| 106 |
+
"language_model.layers.10.self_attn.q_proj",
|
| 107 |
+
"40.self_attn.v_proj",
|
| 108 |
+
"language_model.layers.10.self_attn.v_proj",
|
| 109 |
+
"up_proj",
|
| 110 |
+
"38.self_attn.v_proj",
|
| 111 |
+
"35.self_attn.q_proj",
|
| 112 |
+
"28.self_attn.k_proj",
|
| 113 |
+
"language_model.layers.18.self_attn.k_proj",
|
| 114 |
+
"language_model.layers.26.self_attn.q_proj",
|
| 115 |
+
"38.self_attn.q_proj",
|
| 116 |
+
"36.self_attn.v_proj",
|
| 117 |
+
"language_model.layers.11.self_attn.q_proj",
|
| 118 |
+
"language_model.layers.3.self_attn.k_proj",
|
| 119 |
+
"language_model.layers.24.self_attn.q_proj",
|
| 120 |
+
"37.self_attn.q_proj",
|
| 121 |
+
"45.self_attn.q_proj",
|
| 122 |
+
"42.self_attn.q_proj",
|
| 123 |
+
"language_model.layers.3.self_attn.v_proj",
|
| 124 |
+
"language_model.layers.0.self_attn.v_proj",
|
| 125 |
+
"language_model.layers.23.self_attn.k_proj",
|
| 126 |
+
"language_model.layers.9.self_attn.k_proj",
|
| 127 |
+
"27.self_attn.q_proj",
|
| 128 |
+
"language_model.layers.7.self_attn.q_proj",
|
| 129 |
+
"language_model.layers.25.self_attn.v_proj",
|
| 130 |
+
"30.self_attn.v_proj",
|
| 131 |
+
"language_model.layers.24.self_attn.v_proj",
|
| 132 |
+
"language_model.layers.14.self_attn.v_proj",
|
| 133 |
+
"language_model.layers.21.self_attn.q_proj",
|
| 134 |
+
"46.self_attn.v_proj",
|
| 135 |
+
"language_model.layers.20.self_attn.q_proj",
|
| 136 |
+
"31.self_attn.v_proj",
|
| 137 |
+
"28.self_attn.q_proj",
|
| 138 |
+
"27.self_attn.k_proj",
|
| 139 |
+
"40.self_attn.q_proj",
|
| 140 |
+
"language_model.layers.6.self_attn.k_proj",
|
| 141 |
+
"45.self_attn.k_proj",
|
| 142 |
+
"language_model.layers.1.self_attn.q_proj",
|
| 143 |
+
"38.self_attn.k_proj",
|
| 144 |
+
"language_model.layers.10.self_attn.k_proj",
|
| 145 |
+
"39.self_attn.k_proj",
|
| 146 |
+
"41.self_attn.k_proj",
|
| 147 |
+
"44.self_attn.k_proj",
|
| 148 |
+
"43.self_attn.q_proj",
|
| 149 |
+
"language_model.layers.2.self_attn.k_proj",
|
| 150 |
+
"language_model.layers.22.self_attn.k_proj",
|
| 151 |
+
"language_model.layers.18.self_attn.v_proj",
|
| 152 |
+
"language_model.layers.11.self_attn.v_proj",
|
| 153 |
+
"44.self_attn.q_proj",
|
| 154 |
+
"language_model.layers.2.self_attn.v_proj",
|
| 155 |
+
"down_proj",
|
| 156 |
+
"47.self_attn.v_proj",
|
| 157 |
+
"29.self_attn.v_proj",
|
| 158 |
+
"language_model.layers.21.self_attn.k_proj",
|
| 159 |
+
"35.self_attn.v_proj",
|
| 160 |
+
"46.self_attn.k_proj",
|
| 161 |
+
"language_model.layers.17.self_attn.k_proj",
|
| 162 |
+
"47.self_attn.k_proj",
|
| 163 |
+
"language_model.layers.7.self_attn.v_proj",
|
| 164 |
+
"39.self_attn.q_proj",
|
| 165 |
+
"language_model.layers.26.self_attn.k_proj",
|
| 166 |
+
"language_model.layers.2.self_attn.q_proj",
|
| 167 |
+
"language_model.layers.20.self_attn.v_proj",
|
| 168 |
+
"language_model.layers.14.self_attn.q_proj",
|
| 169 |
+
"language_model.layers.16.self_attn.v_proj",
|
| 170 |
+
"32.self_attn.v_proj",
|
| 171 |
+
"language_model.layers.23.self_attn.v_proj",
|
| 172 |
+
"language_model.layers.19.self_attn.v_proj",
|
| 173 |
+
"27.self_attn.v_proj",
|
| 174 |
+
"language_model.layers.4.self_attn.v_proj",
|
| 175 |
+
"language_model.layers.25.self_attn.q_proj",
|
| 176 |
+
"31.self_attn.q_proj",
|
| 177 |
+
"43.self_attn.k_proj",
|
| 178 |
+
"language_model.layers.4.self_attn.q_proj",
|
| 179 |
+
"language_model.layers.17.self_attn.v_proj",
|
| 180 |
+
"language_model.layers.12.self_attn.v_proj",
|
| 181 |
+
"36.self_attn.k_proj"
|
| 182 |
+
],
|
| 183 |
+
"target_parameters": null,
|
| 184 |
+
"task_type": "CAUSAL_LM",
|
| 185 |
+
"trainable_token_indices": null,
|
| 186 |
+
"use_bdlora": null,
|
| 187 |
+
"use_dora": false,
|
| 188 |
+
"use_qalora": false,
|
| 189 |
+
"use_rslora": false,
|
| 190 |
+
"velora_config": null
|
| 191 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-128/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:44e3d3ffc2775912a205746840bfe6cd2da126f7c59a8715d1d96f848790d757
|
| 3 |
+
size 523864568
|
rl_grpo/coin_real_4x_direct/checkpoint-16/adapter_config.json
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/workspace/rl3_direct_b/parent",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"layer_replication": null,
|
| 16 |
+
"layers_pattern": null,
|
| 17 |
+
"layers_to_transform": null,
|
| 18 |
+
"loftq_config": {},
|
| 19 |
+
"lora_alpha": 64,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 32,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"language_model.layers.13.self_attn.q_proj",
|
| 35 |
+
"46.self_attn.q_proj",
|
| 36 |
+
"39.self_attn.v_proj",
|
| 37 |
+
"language_model.layers.15.self_attn.q_proj",
|
| 38 |
+
"language_model.layers.19.self_attn.k_proj",
|
| 39 |
+
"language_model.layers.21.self_attn.v_proj",
|
| 40 |
+
"language_model.layers.9.self_attn.q_proj",
|
| 41 |
+
"41.self_attn.q_proj",
|
| 42 |
+
"o_proj",
|
| 43 |
+
"language_model.layers.16.self_attn.q_proj",
|
| 44 |
+
"language_model.layers.13.self_attn.k_proj",
|
| 45 |
+
"40.self_attn.k_proj",
|
| 46 |
+
"42.self_attn.k_proj",
|
| 47 |
+
"language_model.layers.18.self_attn.q_proj",
|
| 48 |
+
"language_model.layers.12.self_attn.k_proj",
|
| 49 |
+
"34.self_attn.k_proj",
|
| 50 |
+
"language_model.layers.0.self_attn.q_proj",
|
| 51 |
+
"34.self_attn.q_proj",
|
| 52 |
+
"37.self_attn.v_proj",
|
| 53 |
+
"language_model.layers.12.self_attn.q_proj",
|
| 54 |
+
"33.self_attn.k_proj",
|
| 55 |
+
"language_model.layers.8.self_attn.v_proj",
|
| 56 |
+
"32.self_attn.q_proj",
|
| 57 |
+
"43.self_attn.v_proj",
|
| 58 |
+
"language_model.layers.6.self_attn.q_proj",
|
| 59 |
+
"language_model.layers.7.self_attn.k_proj",
|
| 60 |
+
"language_model.layers.19.self_attn.q_proj",
|
| 61 |
+
"32.self_attn.k_proj",
|
| 62 |
+
"language_model.layers.23.self_attn.q_proj",
|
| 63 |
+
"language_model.layers.3.self_attn.q_proj",
|
| 64 |
+
"language_model.layers.22.self_attn.v_proj",
|
| 65 |
+
"36.self_attn.q_proj",
|
| 66 |
+
"33.self_attn.v_proj",
|
| 67 |
+
"language_model.layers.16.self_attn.k_proj",
|
| 68 |
+
"language_model.layers.1.self_attn.k_proj",
|
| 69 |
+
"language_model.layers.8.self_attn.k_proj",
|
| 70 |
+
"language_model.layers.15.self_attn.k_proj",
|
| 71 |
+
"33.self_attn.q_proj",
|
| 72 |
+
"language_model.layers.15.self_attn.v_proj",
|
| 73 |
+
"language_model.layers.26.self_attn.v_proj",
|
| 74 |
+
"42.self_attn.v_proj",
|
| 75 |
+
"29.self_attn.q_proj",
|
| 76 |
+
"45.self_attn.v_proj",
|
| 77 |
+
"language_model.layers.17.self_attn.q_proj",
|
| 78 |
+
"language_model.layers.24.self_attn.k_proj",
|
| 79 |
+
"language_model.layers.1.self_attn.v_proj",
|
| 80 |
+
"language_model.layers.20.self_attn.k_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"language_model.layers.5.self_attn.q_proj",
|
| 83 |
+
"34.self_attn.v_proj",
|
| 84 |
+
"44.self_attn.v_proj",
|
| 85 |
+
"language_model.layers.0.self_attn.k_proj",
|
| 86 |
+
"28.self_attn.v_proj",
|
| 87 |
+
"language_model.layers.22.self_attn.q_proj",
|
| 88 |
+
"30.self_attn.k_proj",
|
| 89 |
+
"30.self_attn.q_proj",
|
| 90 |
+
"language_model.layers.6.self_attn.v_proj",
|
| 91 |
+
"language_model.layers.14.self_attn.k_proj",
|
| 92 |
+
"language_model.layers.5.self_attn.v_proj",
|
| 93 |
+
"41.self_attn.v_proj",
|
| 94 |
+
"language_model.layers.25.self_attn.k_proj",
|
| 95 |
+
"31.self_attn.k_proj",
|
| 96 |
+
"language_model.layers.5.self_attn.k_proj",
|
| 97 |
+
"language_model.layers.4.self_attn.k_proj",
|
| 98 |
+
"35.self_attn.k_proj",
|
| 99 |
+
"37.self_attn.k_proj",
|
| 100 |
+
"language_model.layers.11.self_attn.k_proj",
|
| 101 |
+
"language_model.layers.8.self_attn.q_proj",
|
| 102 |
+
"29.self_attn.k_proj",
|
| 103 |
+
"47.self_attn.q_proj",
|
| 104 |
+
"language_model.layers.13.self_attn.v_proj",
|
| 105 |
+
"language_model.layers.9.self_attn.v_proj",
|
| 106 |
+
"language_model.layers.10.self_attn.q_proj",
|
| 107 |
+
"40.self_attn.v_proj",
|
| 108 |
+
"language_model.layers.10.self_attn.v_proj",
|
| 109 |
+
"up_proj",
|
| 110 |
+
"38.self_attn.v_proj",
|
| 111 |
+
"35.self_attn.q_proj",
|
| 112 |
+
"28.self_attn.k_proj",
|
| 113 |
+
"language_model.layers.18.self_attn.k_proj",
|
| 114 |
+
"language_model.layers.26.self_attn.q_proj",
|
| 115 |
+
"38.self_attn.q_proj",
|
| 116 |
+
"36.self_attn.v_proj",
|
| 117 |
+
"language_model.layers.11.self_attn.q_proj",
|
| 118 |
+
"language_model.layers.3.self_attn.k_proj",
|
| 119 |
+
"language_model.layers.24.self_attn.q_proj",
|
| 120 |
+
"37.self_attn.q_proj",
|
| 121 |
+
"45.self_attn.q_proj",
|
| 122 |
+
"42.self_attn.q_proj",
|
| 123 |
+
"language_model.layers.3.self_attn.v_proj",
|
| 124 |
+
"language_model.layers.0.self_attn.v_proj",
|
| 125 |
+
"language_model.layers.23.self_attn.k_proj",
|
| 126 |
+
"language_model.layers.9.self_attn.k_proj",
|
| 127 |
+
"27.self_attn.q_proj",
|
| 128 |
+
"language_model.layers.7.self_attn.q_proj",
|
| 129 |
+
"language_model.layers.25.self_attn.v_proj",
|
| 130 |
+
"30.self_attn.v_proj",
|
| 131 |
+
"language_model.layers.24.self_attn.v_proj",
|
| 132 |
+
"language_model.layers.14.self_attn.v_proj",
|
| 133 |
+
"language_model.layers.21.self_attn.q_proj",
|
| 134 |
+
"46.self_attn.v_proj",
|
| 135 |
+
"language_model.layers.20.self_attn.q_proj",
|
| 136 |
+
"31.self_attn.v_proj",
|
| 137 |
+
"28.self_attn.q_proj",
|
| 138 |
+
"27.self_attn.k_proj",
|
| 139 |
+
"40.self_attn.q_proj",
|
| 140 |
+
"language_model.layers.6.self_attn.k_proj",
|
| 141 |
+
"45.self_attn.k_proj",
|
| 142 |
+
"language_model.layers.1.self_attn.q_proj",
|
| 143 |
+
"38.self_attn.k_proj",
|
| 144 |
+
"language_model.layers.10.self_attn.k_proj",
|
| 145 |
+
"39.self_attn.k_proj",
|
| 146 |
+
"41.self_attn.k_proj",
|
| 147 |
+
"44.self_attn.k_proj",
|
| 148 |
+
"43.self_attn.q_proj",
|
| 149 |
+
"language_model.layers.2.self_attn.k_proj",
|
| 150 |
+
"language_model.layers.22.self_attn.k_proj",
|
| 151 |
+
"language_model.layers.18.self_attn.v_proj",
|
| 152 |
+
"language_model.layers.11.self_attn.v_proj",
|
| 153 |
+
"44.self_attn.q_proj",
|
| 154 |
+
"language_model.layers.2.self_attn.v_proj",
|
| 155 |
+
"down_proj",
|
| 156 |
+
"47.self_attn.v_proj",
|
| 157 |
+
"29.self_attn.v_proj",
|
| 158 |
+
"language_model.layers.21.self_attn.k_proj",
|
| 159 |
+
"35.self_attn.v_proj",
|
| 160 |
+
"46.self_attn.k_proj",
|
| 161 |
+
"language_model.layers.17.self_attn.k_proj",
|
| 162 |
+
"47.self_attn.k_proj",
|
| 163 |
+
"language_model.layers.7.self_attn.v_proj",
|
| 164 |
+
"39.self_attn.q_proj",
|
| 165 |
+
"language_model.layers.26.self_attn.k_proj",
|
| 166 |
+
"language_model.layers.2.self_attn.q_proj",
|
| 167 |
+
"language_model.layers.20.self_attn.v_proj",
|
| 168 |
+
"language_model.layers.14.self_attn.q_proj",
|
| 169 |
+
"language_model.layers.16.self_attn.v_proj",
|
| 170 |
+
"32.self_attn.v_proj",
|
| 171 |
+
"language_model.layers.23.self_attn.v_proj",
|
| 172 |
+
"language_model.layers.19.self_attn.v_proj",
|
| 173 |
+
"27.self_attn.v_proj",
|
| 174 |
+
"language_model.layers.4.self_attn.v_proj",
|
| 175 |
+
"language_model.layers.25.self_attn.q_proj",
|
| 176 |
+
"31.self_attn.q_proj",
|
| 177 |
+
"43.self_attn.k_proj",
|
| 178 |
+
"language_model.layers.4.self_attn.q_proj",
|
| 179 |
+
"language_model.layers.17.self_attn.v_proj",
|
| 180 |
+
"language_model.layers.12.self_attn.v_proj",
|
| 181 |
+
"36.self_attn.k_proj"
|
| 182 |
+
],
|
| 183 |
+
"target_parameters": null,
|
| 184 |
+
"task_type": "CAUSAL_LM",
|
| 185 |
+
"trainable_token_indices": null,
|
| 186 |
+
"use_bdlora": null,
|
| 187 |
+
"use_dora": false,
|
| 188 |
+
"use_qalora": false,
|
| 189 |
+
"use_rslora": false,
|
| 190 |
+
"velora_config": null
|
| 191 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-16/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1c3d1ca674dfc58e8ad647bd5312713a2cdbe966ea54dc73c4e18e6d5268b9fe
|
| 3 |
+
size 523864568
|
rl_grpo/coin_real_4x_direct/checkpoint-256/adapter_config.json
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/workspace/rl3_direct_b/parent",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"layer_replication": null,
|
| 16 |
+
"layers_pattern": null,
|
| 17 |
+
"layers_to_transform": null,
|
| 18 |
+
"loftq_config": {},
|
| 19 |
+
"lora_alpha": 64,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 32,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"language_model.layers.13.self_attn.q_proj",
|
| 35 |
+
"46.self_attn.q_proj",
|
| 36 |
+
"39.self_attn.v_proj",
|
| 37 |
+
"language_model.layers.15.self_attn.q_proj",
|
| 38 |
+
"language_model.layers.19.self_attn.k_proj",
|
| 39 |
+
"language_model.layers.21.self_attn.v_proj",
|
| 40 |
+
"language_model.layers.9.self_attn.q_proj",
|
| 41 |
+
"41.self_attn.q_proj",
|
| 42 |
+
"o_proj",
|
| 43 |
+
"language_model.layers.16.self_attn.q_proj",
|
| 44 |
+
"language_model.layers.13.self_attn.k_proj",
|
| 45 |
+
"40.self_attn.k_proj",
|
| 46 |
+
"42.self_attn.k_proj",
|
| 47 |
+
"language_model.layers.18.self_attn.q_proj",
|
| 48 |
+
"language_model.layers.12.self_attn.k_proj",
|
| 49 |
+
"34.self_attn.k_proj",
|
| 50 |
+
"language_model.layers.0.self_attn.q_proj",
|
| 51 |
+
"34.self_attn.q_proj",
|
| 52 |
+
"37.self_attn.v_proj",
|
| 53 |
+
"language_model.layers.12.self_attn.q_proj",
|
| 54 |
+
"33.self_attn.k_proj",
|
| 55 |
+
"language_model.layers.8.self_attn.v_proj",
|
| 56 |
+
"32.self_attn.q_proj",
|
| 57 |
+
"43.self_attn.v_proj",
|
| 58 |
+
"language_model.layers.6.self_attn.q_proj",
|
| 59 |
+
"language_model.layers.7.self_attn.k_proj",
|
| 60 |
+
"language_model.layers.19.self_attn.q_proj",
|
| 61 |
+
"32.self_attn.k_proj",
|
| 62 |
+
"language_model.layers.23.self_attn.q_proj",
|
| 63 |
+
"language_model.layers.3.self_attn.q_proj",
|
| 64 |
+
"language_model.layers.22.self_attn.v_proj",
|
| 65 |
+
"36.self_attn.q_proj",
|
| 66 |
+
"33.self_attn.v_proj",
|
| 67 |
+
"language_model.layers.16.self_attn.k_proj",
|
| 68 |
+
"language_model.layers.1.self_attn.k_proj",
|
| 69 |
+
"language_model.layers.8.self_attn.k_proj",
|
| 70 |
+
"language_model.layers.15.self_attn.k_proj",
|
| 71 |
+
"33.self_attn.q_proj",
|
| 72 |
+
"language_model.layers.15.self_attn.v_proj",
|
| 73 |
+
"language_model.layers.26.self_attn.v_proj",
|
| 74 |
+
"42.self_attn.v_proj",
|
| 75 |
+
"29.self_attn.q_proj",
|
| 76 |
+
"45.self_attn.v_proj",
|
| 77 |
+
"language_model.layers.17.self_attn.q_proj",
|
| 78 |
+
"language_model.layers.24.self_attn.k_proj",
|
| 79 |
+
"language_model.layers.1.self_attn.v_proj",
|
| 80 |
+
"language_model.layers.20.self_attn.k_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"language_model.layers.5.self_attn.q_proj",
|
| 83 |
+
"34.self_attn.v_proj",
|
| 84 |
+
"44.self_attn.v_proj",
|
| 85 |
+
"language_model.layers.0.self_attn.k_proj",
|
| 86 |
+
"28.self_attn.v_proj",
|
| 87 |
+
"language_model.layers.22.self_attn.q_proj",
|
| 88 |
+
"30.self_attn.k_proj",
|
| 89 |
+
"30.self_attn.q_proj",
|
| 90 |
+
"language_model.layers.6.self_attn.v_proj",
|
| 91 |
+
"language_model.layers.14.self_attn.k_proj",
|
| 92 |
+
"language_model.layers.5.self_attn.v_proj",
|
| 93 |
+
"41.self_attn.v_proj",
|
| 94 |
+
"language_model.layers.25.self_attn.k_proj",
|
| 95 |
+
"31.self_attn.k_proj",
|
| 96 |
+
"language_model.layers.5.self_attn.k_proj",
|
| 97 |
+
"language_model.layers.4.self_attn.k_proj",
|
| 98 |
+
"35.self_attn.k_proj",
|
| 99 |
+
"37.self_attn.k_proj",
|
| 100 |
+
"language_model.layers.11.self_attn.k_proj",
|
| 101 |
+
"language_model.layers.8.self_attn.q_proj",
|
| 102 |
+
"29.self_attn.k_proj",
|
| 103 |
+
"47.self_attn.q_proj",
|
| 104 |
+
"language_model.layers.13.self_attn.v_proj",
|
| 105 |
+
"language_model.layers.9.self_attn.v_proj",
|
| 106 |
+
"language_model.layers.10.self_attn.q_proj",
|
| 107 |
+
"40.self_attn.v_proj",
|
| 108 |
+
"language_model.layers.10.self_attn.v_proj",
|
| 109 |
+
"up_proj",
|
| 110 |
+
"38.self_attn.v_proj",
|
| 111 |
+
"35.self_attn.q_proj",
|
| 112 |
+
"28.self_attn.k_proj",
|
| 113 |
+
"language_model.layers.18.self_attn.k_proj",
|
| 114 |
+
"language_model.layers.26.self_attn.q_proj",
|
| 115 |
+
"38.self_attn.q_proj",
|
| 116 |
+
"36.self_attn.v_proj",
|
| 117 |
+
"language_model.layers.11.self_attn.q_proj",
|
| 118 |
+
"language_model.layers.3.self_attn.k_proj",
|
| 119 |
+
"language_model.layers.24.self_attn.q_proj",
|
| 120 |
+
"37.self_attn.q_proj",
|
| 121 |
+
"45.self_attn.q_proj",
|
| 122 |
+
"42.self_attn.q_proj",
|
| 123 |
+
"language_model.layers.3.self_attn.v_proj",
|
| 124 |
+
"language_model.layers.0.self_attn.v_proj",
|
| 125 |
+
"language_model.layers.23.self_attn.k_proj",
|
| 126 |
+
"language_model.layers.9.self_attn.k_proj",
|
| 127 |
+
"27.self_attn.q_proj",
|
| 128 |
+
"language_model.layers.7.self_attn.q_proj",
|
| 129 |
+
"language_model.layers.25.self_attn.v_proj",
|
| 130 |
+
"30.self_attn.v_proj",
|
| 131 |
+
"language_model.layers.24.self_attn.v_proj",
|
| 132 |
+
"language_model.layers.14.self_attn.v_proj",
|
| 133 |
+
"language_model.layers.21.self_attn.q_proj",
|
| 134 |
+
"46.self_attn.v_proj",
|
| 135 |
+
"language_model.layers.20.self_attn.q_proj",
|
| 136 |
+
"31.self_attn.v_proj",
|
| 137 |
+
"28.self_attn.q_proj",
|
| 138 |
+
"27.self_attn.k_proj",
|
| 139 |
+
"40.self_attn.q_proj",
|
| 140 |
+
"language_model.layers.6.self_attn.k_proj",
|
| 141 |
+
"45.self_attn.k_proj",
|
| 142 |
+
"language_model.layers.1.self_attn.q_proj",
|
| 143 |
+
"38.self_attn.k_proj",
|
| 144 |
+
"language_model.layers.10.self_attn.k_proj",
|
| 145 |
+
"39.self_attn.k_proj",
|
| 146 |
+
"41.self_attn.k_proj",
|
| 147 |
+
"44.self_attn.k_proj",
|
| 148 |
+
"43.self_attn.q_proj",
|
| 149 |
+
"language_model.layers.2.self_attn.k_proj",
|
| 150 |
+
"language_model.layers.22.self_attn.k_proj",
|
| 151 |
+
"language_model.layers.18.self_attn.v_proj",
|
| 152 |
+
"language_model.layers.11.self_attn.v_proj",
|
| 153 |
+
"44.self_attn.q_proj",
|
| 154 |
+
"language_model.layers.2.self_attn.v_proj",
|
| 155 |
+
"down_proj",
|
| 156 |
+
"47.self_attn.v_proj",
|
| 157 |
+
"29.self_attn.v_proj",
|
| 158 |
+
"language_model.layers.21.self_attn.k_proj",
|
| 159 |
+
"35.self_attn.v_proj",
|
| 160 |
+
"46.self_attn.k_proj",
|
| 161 |
+
"language_model.layers.17.self_attn.k_proj",
|
| 162 |
+
"47.self_attn.k_proj",
|
| 163 |
+
"language_model.layers.7.self_attn.v_proj",
|
| 164 |
+
"39.self_attn.q_proj",
|
| 165 |
+
"language_model.layers.26.self_attn.k_proj",
|
| 166 |
+
"language_model.layers.2.self_attn.q_proj",
|
| 167 |
+
"language_model.layers.20.self_attn.v_proj",
|
| 168 |
+
"language_model.layers.14.self_attn.q_proj",
|
| 169 |
+
"language_model.layers.16.self_attn.v_proj",
|
| 170 |
+
"32.self_attn.v_proj",
|
| 171 |
+
"language_model.layers.23.self_attn.v_proj",
|
| 172 |
+
"language_model.layers.19.self_attn.v_proj",
|
| 173 |
+
"27.self_attn.v_proj",
|
| 174 |
+
"language_model.layers.4.self_attn.v_proj",
|
| 175 |
+
"language_model.layers.25.self_attn.q_proj",
|
| 176 |
+
"31.self_attn.q_proj",
|
| 177 |
+
"43.self_attn.k_proj",
|
| 178 |
+
"language_model.layers.4.self_attn.q_proj",
|
| 179 |
+
"language_model.layers.17.self_attn.v_proj",
|
| 180 |
+
"language_model.layers.12.self_attn.v_proj",
|
| 181 |
+
"36.self_attn.k_proj"
|
| 182 |
+
],
|
| 183 |
+
"target_parameters": null,
|
| 184 |
+
"task_type": "CAUSAL_LM",
|
| 185 |
+
"trainable_token_indices": null,
|
| 186 |
+
"use_bdlora": null,
|
| 187 |
+
"use_dora": false,
|
| 188 |
+
"use_qalora": false,
|
| 189 |
+
"use_rslora": false,
|
| 190 |
+
"velora_config": null
|
| 191 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-256/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3de438b3de4e15b28d1a2e3f18ef543fac27fcb41760276224f3c8f2f569106f
|
| 3 |
+
size 523864568
|
rl_grpo/coin_real_4x_direct/checkpoint-256/trainer_state.json
ADDED
|
@@ -0,0 +1,834 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.125,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 256,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"clip_ratio/high_max": 0.0,
|
| 14 |
+
"clip_ratio/high_mean": 0.0,
|
| 15 |
+
"clip_ratio/low_mean": 0.0,
|
| 16 |
+
"clip_ratio/low_min": 0.0,
|
| 17 |
+
"clip_ratio/region_mean": 0.0,
|
| 18 |
+
"completions/clipped_ratio": 0.0375,
|
| 19 |
+
"completions/max_length": 100.2,
|
| 20 |
+
"completions/max_terminated_length": 39.275,
|
| 21 |
+
"completions/mean_length": 31.76875,
|
| 22 |
+
"completions/mean_terminated_length": 23.057440757751465,
|
| 23 |
+
"completions/min_length": 19.6,
|
| 24 |
+
"completions/min_terminated_length": 19.6,
|
| 25 |
+
"entropy": 0.18429145347326994,
|
| 26 |
+
"epoch": 0.0048828125,
|
| 27 |
+
"frac_reward_zero_std": 0.15,
|
| 28 |
+
"grad_norm": 0.14375615119934082,
|
| 29 |
+
"learning_rate": 9.648437500000002e-06,
|
| 30 |
+
"loss": 0.00031249811872839927,
|
| 31 |
+
"num_tokens": 225478.0,
|
| 32 |
+
"reward": 0.496875,
|
| 33 |
+
"reward_std": 0.3228072978556156,
|
| 34 |
+
"rewards/reward_func/mean": 0.496875,
|
| 35 |
+
"rewards/reward_func/std": 0.3228072978556156,
|
| 36 |
+
"sampling/importance_sampling_ratio/max": 1.1751019686460495,
|
| 37 |
+
"sampling/importance_sampling_ratio/mean": 1.0005044043064117,
|
| 38 |
+
"sampling/importance_sampling_ratio/min": 0.8161399632692337,
|
| 39 |
+
"sampling/sampling_logp_difference/max": 0.2160791328176856,
|
| 40 |
+
"sampling/sampling_logp_difference/mean": 0.008641275216359646,
|
| 41 |
+
"step": 10,
|
| 42 |
+
"step_time": 15.166030184086413
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"clip_ratio/high_max": 0.0,
|
| 46 |
+
"clip_ratio/high_mean": 0.0,
|
| 47 |
+
"clip_ratio/low_mean": 0.0,
|
| 48 |
+
"clip_ratio/low_min": 0.0,
|
| 49 |
+
"clip_ratio/region_mean": 0.0,
|
| 50 |
+
"completions/clipped_ratio": 0.0375,
|
| 51 |
+
"completions/max_length": 101.35,
|
| 52 |
+
"completions/max_terminated_length": 53.425,
|
| 53 |
+
"completions/mean_length": 34.790625,
|
| 54 |
+
"completions/mean_terminated_length": 26.195833683013916,
|
| 55 |
+
"completions/min_length": 20.75,
|
| 56 |
+
"completions/min_terminated_length": 20.75,
|
| 57 |
+
"entropy": 0.16174604343250393,
|
| 58 |
+
"epoch": 0.009765625,
|
| 59 |
+
"frac_reward_zero_std": 0.1,
|
| 60 |
+
"grad_norm": 0.07719279080629349,
|
| 61 |
+
"learning_rate": 9.2578125e-06,
|
| 62 |
+
"loss": -0.004828216508030891,
|
| 63 |
+
"num_tokens": 465459.0,
|
| 64 |
+
"reward": 0.4875,
|
| 65 |
+
"reward_std": 0.3401901964098215,
|
| 66 |
+
"rewards/reward_func/mean": 0.4875,
|
| 67 |
+
"rewards/reward_func/std": 0.3401901964098215,
|
| 68 |
+
"sampling/importance_sampling_ratio/max": 1.2217658191919327,
|
| 69 |
+
"sampling/importance_sampling_ratio/mean": 0.9938467666506767,
|
| 70 |
+
"sampling/importance_sampling_ratio/min": 0.7148067899048328,
|
| 71 |
+
"sampling/sampling_logp_difference/max": 0.3595921561121941,
|
| 72 |
+
"sampling/sampling_logp_difference/mean": 0.011987832543672994,
|
| 73 |
+
"step": 20,
|
| 74 |
+
"step_time": 15.789622888015582
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"clip_ratio/high_max": 0.0,
|
| 78 |
+
"clip_ratio/high_mean": 0.0,
|
| 79 |
+
"clip_ratio/low_mean": 0.0,
|
| 80 |
+
"clip_ratio/low_min": 0.0,
|
| 81 |
+
"clip_ratio/region_mean": 0.0,
|
| 82 |
+
"completions/clipped_ratio": 0.01875,
|
| 83 |
+
"completions/max_length": 76.225,
|
| 84 |
+
"completions/max_terminated_length": 45.5,
|
| 85 |
+
"completions/mean_length": 28.634375,
|
| 86 |
+
"completions/mean_terminated_length": 24.30758943557739,
|
| 87 |
+
"completions/min_length": 20.525,
|
| 88 |
+
"completions/min_terminated_length": 20.525,
|
| 89 |
+
"entropy": 0.09056603969074786,
|
| 90 |
+
"epoch": 0.0146484375,
|
| 91 |
+
"frac_reward_zero_std": 0.325,
|
| 92 |
+
"grad_norm": 0.2063492089509964,
|
| 93 |
+
"learning_rate": 8.867187500000001e-06,
|
| 94 |
+
"loss": -0.009060102701187133,
|
| 95 |
+
"num_tokens": 694358.0,
|
| 96 |
+
"reward": 0.6890625,
|
| 97 |
+
"reward_std": 0.19485458508133888,
|
| 98 |
+
"rewards/reward_func/mean": 0.6890625,
|
| 99 |
+
"rewards/reward_func/std": 0.19485458508133888,
|
| 100 |
+
"sampling/importance_sampling_ratio/max": 1.2735351771116257,
|
| 101 |
+
"sampling/importance_sampling_ratio/mean": 0.9505834899842739,
|
| 102 |
+
"sampling/importance_sampling_ratio/min": 0.4540431661531329,
|
| 103 |
+
"sampling/sampling_logp_difference/max": 1.0098584877792747,
|
| 104 |
+
"sampling/sampling_logp_difference/mean": 0.020204850373556838,
|
| 105 |
+
"step": 30,
|
| 106 |
+
"step_time": 13.551629175618292
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"clip_ratio/high_max": 0.0,
|
| 110 |
+
"clip_ratio/high_mean": 0.0,
|
| 111 |
+
"clip_ratio/low_mean": 0.0,
|
| 112 |
+
"clip_ratio/low_min": 0.0,
|
| 113 |
+
"clip_ratio/region_mean": 0.0,
|
| 114 |
+
"completions/clipped_ratio": 0.00625,
|
| 115 |
+
"completions/max_length": 42.65,
|
| 116 |
+
"completions/max_terminated_length": 31.05,
|
| 117 |
+
"completions/mean_length": 23.8625,
|
| 118 |
+
"completions/mean_terminated_length": 22.399107217788696,
|
| 119 |
+
"completions/min_length": 20.425,
|
| 120 |
+
"completions/min_terminated_length": 20.425,
|
| 121 |
+
"entropy": 0.051669922564178704,
|
| 122 |
+
"epoch": 0.01953125,
|
| 123 |
+
"frac_reward_zero_std": 0.6,
|
| 124 |
+
"grad_norm": 0.0,
|
| 125 |
+
"learning_rate": 8.4765625e-06,
|
| 126 |
+
"loss": -0.00437207892537117,
|
| 127 |
+
"num_tokens": 917666.0,
|
| 128 |
+
"reward": 0.7984375,
|
| 129 |
+
"reward_std": 0.12761950865387917,
|
| 130 |
+
"rewards/reward_func/mean": 0.7984375,
|
| 131 |
+
"rewards/reward_func/std": 0.12761950865387917,
|
| 132 |
+
"sampling/importance_sampling_ratio/max": 1.2350739628076552,
|
| 133 |
+
"sampling/importance_sampling_ratio/mean": 0.9586559005081654,
|
| 134 |
+
"sampling/importance_sampling_ratio/min": 0.5215567184379324,
|
| 135 |
+
"sampling/sampling_logp_difference/max": 1.0551308968104423,
|
| 136 |
+
"sampling/sampling_logp_difference/mean": 0.018048708559945226,
|
| 137 |
+
"step": 40,
|
| 138 |
+
"step_time": 11.60720629838761
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"clip_ratio/high_max": 0.0,
|
| 142 |
+
"clip_ratio/high_mean": 0.0,
|
| 143 |
+
"clip_ratio/low_mean": 0.00013440860202535988,
|
| 144 |
+
"clip_ratio/low_min": 0.0,
|
| 145 |
+
"clip_ratio/region_mean": 0.00013440860202535988,
|
| 146 |
+
"completions/clipped_ratio": 0.003125,
|
| 147 |
+
"completions/max_length": 32.35,
|
| 148 |
+
"completions/max_terminated_length": 26.55,
|
| 149 |
+
"completions/mean_length": 21.8375,
|
| 150 |
+
"completions/mean_terminated_length": 21.112500047683717,
|
| 151 |
+
"completions/min_length": 20.0,
|
| 152 |
+
"completions/min_terminated_length": 20.0,
|
| 153 |
+
"entropy": 0.04006199312862009,
|
| 154 |
+
"epoch": 0.0244140625,
|
| 155 |
+
"frac_reward_zero_std": 0.55,
|
| 156 |
+
"grad_norm": 0.07198888808488846,
|
| 157 |
+
"learning_rate": 8.085937500000001e-06,
|
| 158 |
+
"loss": -0.007528729736804962,
|
| 159 |
+
"num_tokens": 1141014.0,
|
| 160 |
+
"reward": 0.7421875,
|
| 161 |
+
"reward_std": 0.13416844308376313,
|
| 162 |
+
"rewards/reward_func/mean": 0.7421875,
|
| 163 |
+
"rewards/reward_func/std": 0.13416844308376313,
|
| 164 |
+
"sampling/importance_sampling_ratio/max": 1.1534327648580074,
|
| 165 |
+
"sampling/importance_sampling_ratio/mean": 0.9343447227962315,
|
| 166 |
+
"sampling/importance_sampling_ratio/min": 0.5430467446567491,
|
| 167 |
+
"sampling/sampling_logp_difference/max": 1.2960176127031446,
|
| 168 |
+
"sampling/sampling_logp_difference/mean": 0.020754064549691975,
|
| 169 |
+
"step": 50,
|
| 170 |
+
"step_time": 10.944049425958656
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"clip_ratio/high_max": 0.0,
|
| 174 |
+
"clip_ratio/high_mean": 0.0,
|
| 175 |
+
"clip_ratio/low_mean": 0.0,
|
| 176 |
+
"clip_ratio/low_min": 0.0,
|
| 177 |
+
"clip_ratio/region_mean": 0.0,
|
| 178 |
+
"completions/clipped_ratio": 0.0,
|
| 179 |
+
"completions/max_length": 31.4,
|
| 180 |
+
"completions/max_terminated_length": 31.4,
|
| 181 |
+
"completions/mean_length": 21.915625,
|
| 182 |
+
"completions/mean_terminated_length": 21.915625,
|
| 183 |
+
"completions/min_length": 20.475,
|
| 184 |
+
"completions/min_terminated_length": 20.475,
|
| 185 |
+
"entropy": 0.0365459960186854,
|
| 186 |
+
"epoch": 0.029296875,
|
| 187 |
+
"frac_reward_zero_std": 0.85,
|
| 188 |
+
"grad_norm": 0.04547181352972984,
|
| 189 |
+
"learning_rate": 7.6953125e-06,
|
| 190 |
+
"loss": -0.0017231885343790055,
|
| 191 |
+
"num_tokens": 1366939.0,
|
| 192 |
+
"reward": 0.753125,
|
| 193 |
+
"reward_std": 0.037405286356806754,
|
| 194 |
+
"rewards/reward_func/mean": 0.753125,
|
| 195 |
+
"rewards/reward_func/std": 0.037405286356806754,
|
| 196 |
+
"sampling/importance_sampling_ratio/max": 1.1391231045126915,
|
| 197 |
+
"sampling/importance_sampling_ratio/mean": 1.0052518188953399,
|
| 198 |
+
"sampling/importance_sampling_ratio/min": 0.611104917852208,
|
| 199 |
+
"sampling/sampling_logp_difference/max": 0.6982594026252628,
|
| 200 |
+
"sampling/sampling_logp_difference/mean": 0.012335818688734434,
|
| 201 |
+
"step": 60,
|
| 202 |
+
"step_time": 10.820376485539601
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"clip_ratio/high_max": 0.0,
|
| 206 |
+
"clip_ratio/high_mean": 0.0,
|
| 207 |
+
"clip_ratio/low_mean": 0.0,
|
| 208 |
+
"clip_ratio/low_min": 0.0,
|
| 209 |
+
"clip_ratio/region_mean": 0.0,
|
| 210 |
+
"completions/clipped_ratio": 0.0,
|
| 211 |
+
"completions/max_length": 31.1,
|
| 212 |
+
"completions/max_terminated_length": 31.1,
|
| 213 |
+
"completions/mean_length": 21.459375,
|
| 214 |
+
"completions/mean_terminated_length": 21.459375,
|
| 215 |
+
"completions/min_length": 19.95,
|
| 216 |
+
"completions/min_terminated_length": 19.95,
|
| 217 |
+
"entropy": 0.03331582968821749,
|
| 218 |
+
"epoch": 0.0341796875,
|
| 219 |
+
"frac_reward_zero_std": 0.775,
|
| 220 |
+
"grad_norm": 0.003076166845858097,
|
| 221 |
+
"learning_rate": 7.3046875000000005e-06,
|
| 222 |
+
"loss": -0.0007063730619847775,
|
| 223 |
+
"num_tokens": 1591398.0,
|
| 224 |
+
"reward": 0.8984375,
|
| 225 |
+
"reward_std": 0.07016962952911854,
|
| 226 |
+
"rewards/reward_func/mean": 0.8984375,
|
| 227 |
+
"rewards/reward_func/std": 0.07016962952911854,
|
| 228 |
+
"sampling/importance_sampling_ratio/max": 1.1154227763414384,
|
| 229 |
+
"sampling/importance_sampling_ratio/mean": 0.9720788553357125,
|
| 230 |
+
"sampling/importance_sampling_ratio/min": 0.622283871890977,
|
| 231 |
+
"sampling/sampling_logp_difference/max": 0.6980949867516756,
|
| 232 |
+
"sampling/sampling_logp_difference/mean": 0.014065395573561545,
|
| 233 |
+
"step": 70,
|
| 234 |
+
"step_time": 10.823810588172638
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"clip_ratio/high_max": 0.0,
|
| 238 |
+
"clip_ratio/high_mean": 0.0,
|
| 239 |
+
"clip_ratio/low_mean": 0.0,
|
| 240 |
+
"clip_ratio/low_min": 0.0,
|
| 241 |
+
"clip_ratio/region_mean": 0.0,
|
| 242 |
+
"completions/clipped_ratio": 0.003125,
|
| 243 |
+
"completions/max_length": 29.5,
|
| 244 |
+
"completions/max_terminated_length": 23.55,
|
| 245 |
+
"completions/mean_length": 21.165625,
|
| 246 |
+
"completions/mean_terminated_length": 20.416517877578734,
|
| 247 |
+
"completions/min_length": 19.825,
|
| 248 |
+
"completions/min_terminated_length": 19.825,
|
| 249 |
+
"entropy": 0.02531700908439234,
|
| 250 |
+
"epoch": 0.0390625,
|
| 251 |
+
"frac_reward_zero_std": 0.75,
|
| 252 |
+
"grad_norm": 0.0,
|
| 253 |
+
"learning_rate": 6.9140625e-06,
|
| 254 |
+
"loss": -0.0028342559933662416,
|
| 255 |
+
"num_tokens": 1805875.0,
|
| 256 |
+
"reward": 0.725,
|
| 257 |
+
"reward_std": 0.07827338427305222,
|
| 258 |
+
"rewards/reward_func/mean": 0.725,
|
| 259 |
+
"rewards/reward_func/std": 0.07827338427305222,
|
| 260 |
+
"sampling/importance_sampling_ratio/max": 1.0396482238546014,
|
| 261 |
+
"sampling/importance_sampling_ratio/mean": 0.8946122776484117,
|
| 262 |
+
"sampling/importance_sampling_ratio/min": 0.6213053043698892,
|
| 263 |
+
"sampling/sampling_logp_difference/max": 0.9627589256502688,
|
| 264 |
+
"sampling/sampling_logp_difference/mean": 0.02650494632543996,
|
| 265 |
+
"step": 80,
|
| 266 |
+
"step_time": 10.274059816892258
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"clip_ratio/high_max": 0.0,
|
| 270 |
+
"clip_ratio/high_mean": 0.0,
|
| 271 |
+
"clip_ratio/low_mean": 0.0,
|
| 272 |
+
"clip_ratio/low_min": 0.0,
|
| 273 |
+
"clip_ratio/region_mean": 0.0,
|
| 274 |
+
"completions/clipped_ratio": 0.003125,
|
| 275 |
+
"completions/max_length": 30.675,
|
| 276 |
+
"completions/max_terminated_length": 24.725,
|
| 277 |
+
"completions/mean_length": 22.109375,
|
| 278 |
+
"completions/mean_terminated_length": 21.360267877578735,
|
| 279 |
+
"completions/min_length": 20.8,
|
| 280 |
+
"completions/min_terminated_length": 20.8,
|
| 281 |
+
"entropy": 0.024062751943711193,
|
| 282 |
+
"epoch": 0.0439453125,
|
| 283 |
+
"frac_reward_zero_std": 0.8,
|
| 284 |
+
"grad_norm": 0.04071977734565735,
|
| 285 |
+
"learning_rate": 6.5234375e-06,
|
| 286 |
+
"loss": -0.005896540358662605,
|
| 287 |
+
"num_tokens": 2041902.0,
|
| 288 |
+
"reward": 0.771875,
|
| 289 |
+
"reward_std": 0.06414224654436111,
|
| 290 |
+
"rewards/reward_func/mean": 0.771875,
|
| 291 |
+
"rewards/reward_func/std": 0.06414224654436111,
|
| 292 |
+
"sampling/importance_sampling_ratio/max": 1.113425001622818,
|
| 293 |
+
"sampling/importance_sampling_ratio/mean": 0.9589421071897959,
|
| 294 |
+
"sampling/importance_sampling_ratio/min": 0.6339935208651241,
|
| 295 |
+
"sampling/sampling_logp_difference/max": 1.0300382183864714,
|
| 296 |
+
"sampling/sampling_logp_difference/mean": 0.03200549560424406,
|
| 297 |
+
"step": 90,
|
| 298 |
+
"step_time": 10.996005348931067
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"clip_ratio/high_max": 0.0,
|
| 302 |
+
"clip_ratio/high_mean": 0.0,
|
| 303 |
+
"clip_ratio/low_mean": 0.0,
|
| 304 |
+
"clip_ratio/low_min": 0.0,
|
| 305 |
+
"clip_ratio/region_mean": 0.0,
|
| 306 |
+
"completions/clipped_ratio": 0.00625,
|
| 307 |
+
"completions/max_length": 33.6,
|
| 308 |
+
"completions/max_terminated_length": 21.85,
|
| 309 |
+
"completions/mean_length": 22.328125,
|
| 310 |
+
"completions/mean_terminated_length": 20.85535717010498,
|
| 311 |
+
"completions/min_length": 20.675,
|
| 312 |
+
"completions/min_terminated_length": 20.675,
|
| 313 |
+
"entropy": 0.017737349099479615,
|
| 314 |
+
"epoch": 0.048828125,
|
| 315 |
+
"frac_reward_zero_std": 0.875,
|
| 316 |
+
"grad_norm": 0.0,
|
| 317 |
+
"learning_rate": 6.132812500000001e-06,
|
| 318 |
+
"loss": -0.00459984615445137,
|
| 319 |
+
"num_tokens": 2270055.0,
|
| 320 |
+
"reward": 0.7890625,
|
| 321 |
+
"reward_std": 0.028778617456555366,
|
| 322 |
+
"rewards/reward_func/mean": 0.7890625,
|
| 323 |
+
"rewards/reward_func/std": 0.028778617456555366,
|
| 324 |
+
"sampling/importance_sampling_ratio/max": 1.09726552516222,
|
| 325 |
+
"sampling/importance_sampling_ratio/mean": 0.9675815872848034,
|
| 326 |
+
"sampling/importance_sampling_ratio/min": 0.6969301760196686,
|
| 327 |
+
"sampling/sampling_logp_difference/max": 0.6540146941319108,
|
| 328 |
+
"sampling/sampling_logp_difference/mean": 0.011544473981484771,
|
| 329 |
+
"step": 100,
|
| 330 |
+
"step_time": 11.018028863705695
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"clip_ratio/high_max": 0.0,
|
| 334 |
+
"clip_ratio/high_mean": 0.0,
|
| 335 |
+
"clip_ratio/low_mean": 0.0,
|
| 336 |
+
"clip_ratio/low_min": 0.0,
|
| 337 |
+
"clip_ratio/region_mean": 0.0,
|
| 338 |
+
"completions/clipped_ratio": 0.0,
|
| 339 |
+
"completions/max_length": 22.45,
|
| 340 |
+
"completions/max_terminated_length": 22.45,
|
| 341 |
+
"completions/mean_length": 20.984375,
|
| 342 |
+
"completions/mean_terminated_length": 20.984375,
|
| 343 |
+
"completions/min_length": 20.675,
|
| 344 |
+
"completions/min_terminated_length": 20.675,
|
| 345 |
+
"entropy": 0.023635415046010168,
|
| 346 |
+
"epoch": 0.0537109375,
|
| 347 |
+
"frac_reward_zero_std": 0.75,
|
| 348 |
+
"grad_norm": 0.38097792863845825,
|
| 349 |
+
"learning_rate": 5.742187500000001e-06,
|
| 350 |
+
"loss": 0.0011273769661784172,
|
| 351 |
+
"num_tokens": 2496682.0,
|
| 352 |
+
"reward": 0.7796875,
|
| 353 |
+
"reward_std": 0.069752112403512,
|
| 354 |
+
"rewards/reward_func/mean": 0.7796875,
|
| 355 |
+
"rewards/reward_func/std": 0.069752112403512,
|
| 356 |
+
"sampling/importance_sampling_ratio/max": 1.0827945992350578,
|
| 357 |
+
"sampling/importance_sampling_ratio/mean": 0.965285100787878,
|
| 358 |
+
"sampling/importance_sampling_ratio/min": 0.7070282144472003,
|
| 359 |
+
"sampling/sampling_logp_difference/max": 0.6405853286385537,
|
| 360 |
+
"sampling/sampling_logp_difference/mean": 0.013487762739532626,
|
| 361 |
+
"step": 110,
|
| 362 |
+
"step_time": 10.076650768192485
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"clip_ratio/high_max": 0.0,
|
| 366 |
+
"clip_ratio/high_mean": 0.0,
|
| 367 |
+
"clip_ratio/low_mean": 0.0,
|
| 368 |
+
"clip_ratio/low_min": 0.0,
|
| 369 |
+
"clip_ratio/region_mean": 0.0,
|
| 370 |
+
"completions/clipped_ratio": 0.0,
|
| 371 |
+
"completions/max_length": 22.175,
|
| 372 |
+
"completions/max_terminated_length": 22.175,
|
| 373 |
+
"completions/mean_length": 19.50625,
|
| 374 |
+
"completions/mean_terminated_length": 19.50625,
|
| 375 |
+
"completions/min_length": 19.05,
|
| 376 |
+
"completions/min_terminated_length": 19.05,
|
| 377 |
+
"entropy": 0.02276796295773238,
|
| 378 |
+
"epoch": 0.05859375,
|
| 379 |
+
"frac_reward_zero_std": 0.825,
|
| 380 |
+
"grad_norm": 0.026348156854510307,
|
| 381 |
+
"learning_rate": 5.3515625000000005e-06,
|
| 382 |
+
"loss": -0.0008644983172416687,
|
| 383 |
+
"num_tokens": 2696444.0,
|
| 384 |
+
"reward": 0.746875,
|
| 385 |
+
"reward_std": 0.056707005575299264,
|
| 386 |
+
"rewards/reward_func/mean": 0.746875,
|
| 387 |
+
"rewards/reward_func/std": 0.056707005575299264,
|
| 388 |
+
"sampling/importance_sampling_ratio/max": 1.0637279754271731,
|
| 389 |
+
"sampling/importance_sampling_ratio/mean": 0.9651448504067958,
|
| 390 |
+
"sampling/importance_sampling_ratio/min": 0.7294952804106287,
|
| 391 |
+
"sampling/sampling_logp_difference/max": 0.7509342918172479,
|
| 392 |
+
"sampling/sampling_logp_difference/mean": 0.016977779829176143,
|
| 393 |
+
"step": 120,
|
| 394 |
+
"step_time": 9.136625357344746
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"clip_ratio/high_max": 0.0,
|
| 398 |
+
"clip_ratio/high_mean": 0.0,
|
| 399 |
+
"clip_ratio/low_mean": 0.0,
|
| 400 |
+
"clip_ratio/low_min": 0.0,
|
| 401 |
+
"clip_ratio/region_mean": 0.0,
|
| 402 |
+
"completions/clipped_ratio": 0.0,
|
| 403 |
+
"completions/max_length": 20.375,
|
| 404 |
+
"completions/max_terminated_length": 20.375,
|
| 405 |
+
"completions/mean_length": 19.590625,
|
| 406 |
+
"completions/mean_terminated_length": 19.590625,
|
| 407 |
+
"completions/min_length": 19.4,
|
| 408 |
+
"completions/min_terminated_length": 19.4,
|
| 409 |
+
"entropy": 0.018625556805636732,
|
| 410 |
+
"epoch": 0.0634765625,
|
| 411 |
+
"frac_reward_zero_std": 0.85,
|
| 412 |
+
"grad_norm": 0.19615943729877472,
|
| 413 |
+
"learning_rate": 4.9609375e-06,
|
| 414 |
+
"loss": -0.0007678268477320672,
|
| 415 |
+
"num_tokens": 2907809.0,
|
| 416 |
+
"reward": 0.703125,
|
| 417 |
+
"reward_std": 0.04171832725405693,
|
| 418 |
+
"rewards/reward_func/mean": 0.703125,
|
| 419 |
+
"rewards/reward_func/std": 0.04171832725405693,
|
| 420 |
+
"sampling/importance_sampling_ratio/max": 1.11006118953228,
|
| 421 |
+
"sampling/importance_sampling_ratio/mean": 0.9953158229589463,
|
| 422 |
+
"sampling/importance_sampling_ratio/min": 0.7512783714555553,
|
| 423 |
+
"sampling/sampling_logp_difference/max": 0.5867231288924813,
|
| 424 |
+
"sampling/sampling_logp_difference/mean": 0.01064404183125589,
|
| 425 |
+
"step": 130,
|
| 426 |
+
"step_time": 9.47109698774293
|
| 427 |
+
},
|
| 428 |
+
{
|
| 429 |
+
"clip_ratio/high_max": 0.0,
|
| 430 |
+
"clip_ratio/high_mean": 0.0,
|
| 431 |
+
"clip_ratio/low_mean": 0.0,
|
| 432 |
+
"clip_ratio/low_min": 0.0,
|
| 433 |
+
"clip_ratio/region_mean": 0.0,
|
| 434 |
+
"completions/clipped_ratio": 0.003125,
|
| 435 |
+
"completions/max_length": 26.55,
|
| 436 |
+
"completions/max_terminated_length": 20.8,
|
| 437 |
+
"completions/mean_length": 20.778125,
|
| 438 |
+
"completions/mean_terminated_length": 20.054017877578737,
|
| 439 |
+
"completions/min_length": 19.9,
|
| 440 |
+
"completions/min_terminated_length": 19.9,
|
| 441 |
+
"entropy": 0.018922946101520212,
|
| 442 |
+
"epoch": 0.068359375,
|
| 443 |
+
"frac_reward_zero_std": 0.8,
|
| 444 |
+
"grad_norm": 0.0,
|
| 445 |
+
"learning_rate": 4.5703125e-06,
|
| 446 |
+
"loss": -0.004512551799416542,
|
| 447 |
+
"num_tokens": 3128850.0,
|
| 448 |
+
"reward": 0.85,
|
| 449 |
+
"reward_std": 0.04613676518201828,
|
| 450 |
+
"rewards/reward_func/mean": 0.85,
|
| 451 |
+
"rewards/reward_func/std": 0.04613676518201828,
|
| 452 |
+
"sampling/importance_sampling_ratio/max": 1.0751549154520035,
|
| 453 |
+
"sampling/importance_sampling_ratio/mean": 0.98668732047081,
|
| 454 |
+
"sampling/importance_sampling_ratio/min": 0.7128875293768943,
|
| 455 |
+
"sampling/sampling_logp_difference/max": 0.5808297571726143,
|
| 456 |
+
"sampling/sampling_logp_difference/mean": 0.008342754216573666,
|
| 457 |
+
"step": 140,
|
| 458 |
+
"step_time": 10.109641767875292
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"clip_ratio/high_max": 0.0,
|
| 462 |
+
"clip_ratio/high_mean": 0.0,
|
| 463 |
+
"clip_ratio/low_mean": 0.0,
|
| 464 |
+
"clip_ratio/low_min": 0.0,
|
| 465 |
+
"clip_ratio/region_mean": 0.0,
|
| 466 |
+
"completions/clipped_ratio": 0.0,
|
| 467 |
+
"completions/max_length": 27.05,
|
| 468 |
+
"completions/max_terminated_length": 27.05,
|
| 469 |
+
"completions/mean_length": 21.440625,
|
| 470 |
+
"completions/mean_terminated_length": 21.440625,
|
| 471 |
+
"completions/min_length": 20.575,
|
| 472 |
+
"completions/min_terminated_length": 20.575,
|
| 473 |
+
"entropy": 0.02567918471759185,
|
| 474 |
+
"epoch": 0.0732421875,
|
| 475 |
+
"frac_reward_zero_std": 0.65,
|
| 476 |
+
"grad_norm": 0.4445314109325409,
|
| 477 |
+
"learning_rate": 4.1796875e-06,
|
| 478 |
+
"loss": 0.0020151712000370027,
|
| 479 |
+
"num_tokens": 3354263.0,
|
| 480 |
+
"reward": 0.8078125,
|
| 481 |
+
"reward_std": 0.1015449870377779,
|
| 482 |
+
"rewards/reward_func/mean": 0.8078125,
|
| 483 |
+
"rewards/reward_func/std": 0.1015449870377779,
|
| 484 |
+
"sampling/importance_sampling_ratio/max": 1.1336258022987749,
|
| 485 |
+
"sampling/importance_sampling_ratio/mean": 0.9686125831649406,
|
| 486 |
+
"sampling/importance_sampling_ratio/min": 0.6582718021876645,
|
| 487 |
+
"sampling/sampling_logp_difference/max": 0.9255828734487295,
|
| 488 |
+
"sampling/sampling_logp_difference/mean": 0.017220850277226418,
|
| 489 |
+
"step": 150,
|
| 490 |
+
"step_time": 10.467305223527365
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"clip_ratio/high_max": 0.0,
|
| 494 |
+
"clip_ratio/high_mean": 0.0,
|
| 495 |
+
"clip_ratio/low_mean": 0.0,
|
| 496 |
+
"clip_ratio/low_min": 0.0,
|
| 497 |
+
"clip_ratio/region_mean": 0.0,
|
| 498 |
+
"completions/clipped_ratio": 0.0,
|
| 499 |
+
"completions/max_length": 21.55,
|
| 500 |
+
"completions/max_terminated_length": 21.55,
|
| 501 |
+
"completions/mean_length": 20.5875,
|
| 502 |
+
"completions/mean_terminated_length": 20.5875,
|
| 503 |
+
"completions/min_length": 20.375,
|
| 504 |
+
"completions/min_terminated_length": 20.375,
|
| 505 |
+
"entropy": 0.02475966952042654,
|
| 506 |
+
"epoch": 0.078125,
|
| 507 |
+
"frac_reward_zero_std": 0.725,
|
| 508 |
+
"grad_norm": 0.009551016613841057,
|
| 509 |
+
"learning_rate": 3.7890625e-06,
|
| 510 |
+
"loss": 0.00027390196919441223,
|
| 511 |
+
"num_tokens": 3576843.0,
|
| 512 |
+
"reward": 0.7796875,
|
| 513 |
+
"reward_std": 0.06870663426816463,
|
| 514 |
+
"rewards/reward_func/mean": 0.7796875,
|
| 515 |
+
"rewards/reward_func/std": 0.06870663426816463,
|
| 516 |
+
"sampling/importance_sampling_ratio/max": 1.124920417740941,
|
| 517 |
+
"sampling/importance_sampling_ratio/mean": 0.9702759532257914,
|
| 518 |
+
"sampling/importance_sampling_ratio/min": 0.6678009340539575,
|
| 519 |
+
"sampling/sampling_logp_difference/max": 0.6452897194772959,
|
| 520 |
+
"sampling/sampling_logp_difference/mean": 0.014041122866910883,
|
| 521 |
+
"step": 160,
|
| 522 |
+
"step_time": 10.05609206277877
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"clip_ratio/high_max": 0.0,
|
| 526 |
+
"clip_ratio/high_mean": 0.0,
|
| 527 |
+
"clip_ratio/low_mean": 0.0,
|
| 528 |
+
"clip_ratio/low_min": 0.0,
|
| 529 |
+
"clip_ratio/region_mean": 0.0,
|
| 530 |
+
"completions/clipped_ratio": 0.0,
|
| 531 |
+
"completions/max_length": 31.475,
|
| 532 |
+
"completions/max_terminated_length": 31.475,
|
| 533 |
+
"completions/mean_length": 21.975,
|
| 534 |
+
"completions/mean_terminated_length": 21.975,
|
| 535 |
+
"completions/min_length": 20.5,
|
| 536 |
+
"completions/min_terminated_length": 20.5,
|
| 537 |
+
"entropy": 0.034581871947739276,
|
| 538 |
+
"epoch": 0.0830078125,
|
| 539 |
+
"frac_reward_zero_std": 0.85,
|
| 540 |
+
"grad_norm": 0.05165421590209007,
|
| 541 |
+
"learning_rate": 3.3984375000000004e-06,
|
| 542 |
+
"loss": -0.004046498984098435,
|
| 543 |
+
"num_tokens": 3804123.0,
|
| 544 |
+
"reward": 0.7734375,
|
| 545 |
+
"reward_std": 0.041185660660266875,
|
| 546 |
+
"rewards/reward_func/mean": 0.7734375,
|
| 547 |
+
"rewards/reward_func/std": 0.041185660660266875,
|
| 548 |
+
"sampling/importance_sampling_ratio/max": 1.126641868799925,
|
| 549 |
+
"sampling/importance_sampling_ratio/mean": 0.9788845136761666,
|
| 550 |
+
"sampling/importance_sampling_ratio/min": 0.6990331819746644,
|
| 551 |
+
"sampling/sampling_logp_difference/max": 0.5651584011502564,
|
| 552 |
+
"sampling/sampling_logp_difference/mean": 0.014526259174454026,
|
| 553 |
+
"step": 170,
|
| 554 |
+
"step_time": 10.983829106041231
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"clip_ratio/high_max": 0.0,
|
| 558 |
+
"clip_ratio/high_mean": 0.0,
|
| 559 |
+
"clip_ratio/low_mean": 0.0,
|
| 560 |
+
"clip_ratio/low_min": 0.0,
|
| 561 |
+
"clip_ratio/region_mean": 0.0,
|
| 562 |
+
"completions/clipped_ratio": 0.003125,
|
| 563 |
+
"completions/max_length": 30.675,
|
| 564 |
+
"completions/max_terminated_length": 24.725,
|
| 565 |
+
"completions/mean_length": 22.090625,
|
| 566 |
+
"completions/mean_terminated_length": 21.341517877578735,
|
| 567 |
+
"completions/min_length": 20.675,
|
| 568 |
+
"completions/min_terminated_length": 20.675,
|
| 569 |
+
"entropy": 0.036319971387274566,
|
| 570 |
+
"epoch": 0.087890625,
|
| 571 |
+
"frac_reward_zero_std": 0.725,
|
| 572 |
+
"grad_norm": 0.0,
|
| 573 |
+
"learning_rate": 3.0078125000000003e-06,
|
| 574 |
+
"loss": -0.002558111771941185,
|
| 575 |
+
"num_tokens": 4028312.0,
|
| 576 |
+
"reward": 0.7734375,
|
| 577 |
+
"reward_std": 0.07607017531991005,
|
| 578 |
+
"rewards/reward_func/mean": 0.7734375,
|
| 579 |
+
"rewards/reward_func/std": 0.07607017531991005,
|
| 580 |
+
"sampling/importance_sampling_ratio/max": 1.0801800325512887,
|
| 581 |
+
"sampling/importance_sampling_ratio/mean": 0.9781170785427094,
|
| 582 |
+
"sampling/importance_sampling_ratio/min": 0.6795335489790887,
|
| 583 |
+
"sampling/sampling_logp_difference/max": 0.8411830332130193,
|
| 584 |
+
"sampling/sampling_logp_difference/mean": 0.011184986798616592,
|
| 585 |
+
"step": 180,
|
| 586 |
+
"step_time": 10.711151289613918
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"clip_ratio/high_max": 0.0,
|
| 590 |
+
"clip_ratio/high_mean": 0.0,
|
| 591 |
+
"clip_ratio/low_mean": 0.0,
|
| 592 |
+
"clip_ratio/low_min": 0.0,
|
| 593 |
+
"clip_ratio/region_mean": 0.0,
|
| 594 |
+
"completions/clipped_ratio": 0.003125,
|
| 595 |
+
"completions/max_length": 27.8,
|
| 596 |
+
"completions/max_terminated_length": 23.45,
|
| 597 |
+
"completions/mean_length": 21.6625,
|
| 598 |
+
"completions/mean_terminated_length": 20.966964340209962,
|
| 599 |
+
"completions/min_length": 20.4,
|
| 600 |
+
"completions/min_terminated_length": 20.4,
|
| 601 |
+
"entropy": 0.035199500853195786,
|
| 602 |
+
"epoch": 0.0927734375,
|
| 603 |
+
"frac_reward_zero_std": 0.85,
|
| 604 |
+
"grad_norm": 0.036633942276239395,
|
| 605 |
+
"learning_rate": 2.6171875e-06,
|
| 606 |
+
"loss": -0.003652573376893997,
|
| 607 |
+
"num_tokens": 4258036.0,
|
| 608 |
+
"reward": 0.8734375,
|
| 609 |
+
"reward_std": 0.0506619717925787,
|
| 610 |
+
"rewards/reward_func/mean": 0.8734375,
|
| 611 |
+
"rewards/reward_func/std": 0.0506619717925787,
|
| 612 |
+
"sampling/importance_sampling_ratio/max": 1.0321827258914709,
|
| 613 |
+
"sampling/importance_sampling_ratio/mean": 0.9498035926371813,
|
| 614 |
+
"sampling/importance_sampling_ratio/min": 0.7055012997006997,
|
| 615 |
+
"sampling/sampling_logp_difference/max": 0.6118156844750047,
|
| 616 |
+
"sampling/sampling_logp_difference/mean": 0.012727440425078385,
|
| 617 |
+
"step": 190,
|
| 618 |
+
"step_time": 10.902317494829186
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"clip_ratio/high_max": 0.0,
|
| 622 |
+
"clip_ratio/high_mean": 0.0,
|
| 623 |
+
"clip_ratio/low_mean": 0.0,
|
| 624 |
+
"clip_ratio/low_min": 0.0,
|
| 625 |
+
"clip_ratio/region_mean": 0.0,
|
| 626 |
+
"completions/clipped_ratio": 0.0,
|
| 627 |
+
"completions/max_length": 23.55,
|
| 628 |
+
"completions/max_terminated_length": 23.55,
|
| 629 |
+
"completions/mean_length": 20.984375,
|
| 630 |
+
"completions/mean_terminated_length": 20.984375,
|
| 631 |
+
"completions/min_length": 20.525,
|
| 632 |
+
"completions/min_terminated_length": 20.525,
|
| 633 |
+
"entropy": 0.033642921666614714,
|
| 634 |
+
"epoch": 0.09765625,
|
| 635 |
+
"frac_reward_zero_std": 0.775,
|
| 636 |
+
"grad_norm": 0.0,
|
| 637 |
+
"learning_rate": 2.2265625000000003e-06,
|
| 638 |
+
"loss": -0.005803444981575012,
|
| 639 |
+
"num_tokens": 4488935.0,
|
| 640 |
+
"reward": 0.8640625,
|
| 641 |
+
"reward_std": 0.06428623758256435,
|
| 642 |
+
"rewards/reward_func/mean": 0.8640625,
|
| 643 |
+
"rewards/reward_func/std": 0.06428623758256435,
|
| 644 |
+
"sampling/importance_sampling_ratio/max": 1.104056604206562,
|
| 645 |
+
"sampling/importance_sampling_ratio/mean": 1.0095478639006614,
|
| 646 |
+
"sampling/importance_sampling_ratio/min": 0.7015162892232183,
|
| 647 |
+
"sampling/sampling_logp_difference/max": 0.7781496333424002,
|
| 648 |
+
"sampling/sampling_logp_difference/mean": 0.010512121293868405,
|
| 649 |
+
"step": 200,
|
| 650 |
+
"step_time": 10.731312023755162
|
| 651 |
+
},
|
| 652 |
+
{
|
| 653 |
+
"clip_ratio/high_max": 0.0,
|
| 654 |
+
"clip_ratio/high_mean": 0.0,
|
| 655 |
+
"clip_ratio/low_mean": 0.0,
|
| 656 |
+
"clip_ratio/low_min": 0.0,
|
| 657 |
+
"clip_ratio/region_mean": 0.0,
|
| 658 |
+
"completions/clipped_ratio": 0.00625,
|
| 659 |
+
"completions/max_length": 33.45,
|
| 660 |
+
"completions/max_terminated_length": 21.45,
|
| 661 |
+
"completions/mean_length": 21.753125,
|
| 662 |
+
"completions/mean_terminated_length": 20.253125,
|
| 663 |
+
"completions/min_length": 19.9,
|
| 664 |
+
"completions/min_terminated_length": 19.9,
|
| 665 |
+
"entropy": 0.03452314033638686,
|
| 666 |
+
"epoch": 0.1025390625,
|
| 667 |
+
"frac_reward_zero_std": 0.725,
|
| 668 |
+
"grad_norm": 0.3328929841518402,
|
| 669 |
+
"learning_rate": 1.8359375000000002e-06,
|
| 670 |
+
"loss": -0.00490708202123642,
|
| 671 |
+
"num_tokens": 4712256.0,
|
| 672 |
+
"reward": 0.771875,
|
| 673 |
+
"reward_std": 0.07510230764746666,
|
| 674 |
+
"rewards/reward_func/mean": 0.771875,
|
| 675 |
+
"rewards/reward_func/std": 0.07510230764746666,
|
| 676 |
+
"sampling/importance_sampling_ratio/max": 1.0867192722856998,
|
| 677 |
+
"sampling/importance_sampling_ratio/mean": 0.9542400814592839,
|
| 678 |
+
"sampling/importance_sampling_ratio/min": 0.6727492686826736,
|
| 679 |
+
"sampling/sampling_logp_difference/max": 0.6745983049273491,
|
| 680 |
+
"sampling/sampling_logp_difference/mean": 0.011559607972594676,
|
| 681 |
+
"step": 210,
|
| 682 |
+
"step_time": 10.917276552529074
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"clip_ratio/high_max": 0.0,
|
| 686 |
+
"clip_ratio/high_mean": 0.0,
|
| 687 |
+
"clip_ratio/low_mean": 0.0,
|
| 688 |
+
"clip_ratio/low_min": 0.0,
|
| 689 |
+
"clip_ratio/region_mean": 0.0,
|
| 690 |
+
"completions/clipped_ratio": 0.003125,
|
| 691 |
+
"completions/max_length": 28.475,
|
| 692 |
+
"completions/max_terminated_length": 22.725,
|
| 693 |
+
"completions/mean_length": 22.284375,
|
| 694 |
+
"completions/mean_terminated_length": 21.560267877578735,
|
| 695 |
+
"completions/min_length": 21.2,
|
| 696 |
+
"completions/min_terminated_length": 21.2,
|
| 697 |
+
"entropy": 0.026034280005842448,
|
| 698 |
+
"epoch": 0.107421875,
|
| 699 |
+
"frac_reward_zero_std": 0.825,
|
| 700 |
+
"grad_norm": 0.01493086013942957,
|
| 701 |
+
"learning_rate": 1.4453125000000002e-06,
|
| 702 |
+
"loss": -0.0058102793991565704,
|
| 703 |
+
"num_tokens": 4951619.0,
|
| 704 |
+
"reward": 0.9,
|
| 705 |
+
"reward_std": 0.052030496671795844,
|
| 706 |
+
"rewards/reward_func/mean": 0.9,
|
| 707 |
+
"rewards/reward_func/std": 0.052030496671795844,
|
| 708 |
+
"sampling/importance_sampling_ratio/max": 1.089261144399643,
|
| 709 |
+
"sampling/importance_sampling_ratio/mean": 0.9907616257667542,
|
| 710 |
+
"sampling/importance_sampling_ratio/min": 0.7370164816849865,
|
| 711 |
+
"sampling/sampling_logp_difference/max": 0.7364764382131398,
|
| 712 |
+
"sampling/sampling_logp_difference/mean": 0.010071893414715306,
|
| 713 |
+
"step": 220,
|
| 714 |
+
"step_time": 11.204956495203078
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"clip_ratio/high_max": 0.0,
|
| 718 |
+
"clip_ratio/high_mean": 0.0,
|
| 719 |
+
"clip_ratio/low_mean": 0.0,
|
| 720 |
+
"clip_ratio/low_min": 0.0,
|
| 721 |
+
"clip_ratio/region_mean": 0.0,
|
| 722 |
+
"completions/clipped_ratio": 0.0,
|
| 723 |
+
"completions/max_length": 26.125,
|
| 724 |
+
"completions/max_terminated_length": 26.125,
|
| 725 |
+
"completions/mean_length": 19.85625,
|
| 726 |
+
"completions/mean_terminated_length": 19.85625,
|
| 727 |
+
"completions/min_length": 18.8,
|
| 728 |
+
"completions/min_terminated_length": 18.8,
|
| 729 |
+
"entropy": 0.030222409591078758,
|
| 730 |
+
"epoch": 0.1123046875,
|
| 731 |
+
"frac_reward_zero_std": 0.75,
|
| 732 |
+
"grad_norm": 0.3764279782772064,
|
| 733 |
+
"learning_rate": 1.0546875e-06,
|
| 734 |
+
"loss": -0.008652322739362717,
|
| 735 |
+
"num_tokens": 5153869.0,
|
| 736 |
+
"reward": 0.846875,
|
| 737 |
+
"reward_std": 0.05975946448743343,
|
| 738 |
+
"rewards/reward_func/mean": 0.846875,
|
| 739 |
+
"rewards/reward_func/std": 0.05975946448743343,
|
| 740 |
+
"sampling/importance_sampling_ratio/max": 1.1840050276368856,
|
| 741 |
+
"sampling/importance_sampling_ratio/mean": 0.9419653584249318,
|
| 742 |
+
"sampling/importance_sampling_ratio/min": 0.6662579817464576,
|
| 743 |
+
"sampling/sampling_logp_difference/max": 0.9875782642513513,
|
| 744 |
+
"sampling/sampling_logp_difference/mean": 0.022455378672748338,
|
| 745 |
+
"step": 230,
|
| 746 |
+
"step_time": 9.73484854954295
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"clip_ratio/high_max": 0.0,
|
| 750 |
+
"clip_ratio/high_mean": 0.0,
|
| 751 |
+
"clip_ratio/low_mean": 0.0,
|
| 752 |
+
"clip_ratio/low_min": 0.0,
|
| 753 |
+
"clip_ratio/region_mean": 0.0,
|
| 754 |
+
"completions/clipped_ratio": 0.0,
|
| 755 |
+
"completions/max_length": 25.175,
|
| 756 |
+
"completions/max_terminated_length": 25.175,
|
| 757 |
+
"completions/mean_length": 21.89375,
|
| 758 |
+
"completions/mean_terminated_length": 21.89375,
|
| 759 |
+
"completions/min_length": 21.225,
|
| 760 |
+
"completions/min_terminated_length": 21.225,
|
| 761 |
+
"entropy": 0.03204755829647184,
|
| 762 |
+
"epoch": 0.1171875,
|
| 763 |
+
"frac_reward_zero_std": 0.775,
|
| 764 |
+
"grad_norm": 0.7806635499000549,
|
| 765 |
+
"learning_rate": 6.640625e-07,
|
| 766 |
+
"loss": -0.0015525728464126586,
|
| 767 |
+
"num_tokens": 5391715.0,
|
| 768 |
+
"reward": 0.875,
|
| 769 |
+
"reward_std": 0.06244845539331436,
|
| 770 |
+
"rewards/reward_func/mean": 0.875,
|
| 771 |
+
"rewards/reward_func/std": 0.06244845539331436,
|
| 772 |
+
"sampling/importance_sampling_ratio/max": 1.006112952530384,
|
| 773 |
+
"sampling/importance_sampling_ratio/mean": 0.9516554615460336,
|
| 774 |
+
"sampling/importance_sampling_ratio/min": 0.7286337835714221,
|
| 775 |
+
"sampling/sampling_logp_difference/max": 0.45983982272446156,
|
| 776 |
+
"sampling/sampling_logp_difference/mean": 0.008786546676128637,
|
| 777 |
+
"step": 240,
|
| 778 |
+
"step_time": 10.861664683767595
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"clip_ratio/high_max": 0.0,
|
| 782 |
+
"clip_ratio/high_mean": 0.0,
|
| 783 |
+
"clip_ratio/low_mean": 0.0,
|
| 784 |
+
"clip_ratio/low_min": 0.0,
|
| 785 |
+
"clip_ratio/region_mean": 0.0,
|
| 786 |
+
"completions/clipped_ratio": 0.0,
|
| 787 |
+
"completions/max_length": 24.825,
|
| 788 |
+
"completions/max_terminated_length": 24.825,
|
| 789 |
+
"completions/mean_length": 20.521875,
|
| 790 |
+
"completions/mean_terminated_length": 20.521875,
|
| 791 |
+
"completions/min_length": 19.8,
|
| 792 |
+
"completions/min_terminated_length": 19.8,
|
| 793 |
+
"entropy": 0.030372858396731317,
|
| 794 |
+
"epoch": 0.1220703125,
|
| 795 |
+
"frac_reward_zero_std": 0.8,
|
| 796 |
+
"grad_norm": 0.0,
|
| 797 |
+
"learning_rate": 2.7343750000000003e-07,
|
| 798 |
+
"loss": 0.004847610369324684,
|
| 799 |
+
"num_tokens": 5611178.0,
|
| 800 |
+
"reward": 0.8390625,
|
| 801 |
+
"reward_std": 0.049661027267575264,
|
| 802 |
+
"rewards/reward_func/mean": 0.8390625,
|
| 803 |
+
"rewards/reward_func/std": 0.049661027267575264,
|
| 804 |
+
"sampling/importance_sampling_ratio/max": 1.1019318789243697,
|
| 805 |
+
"sampling/importance_sampling_ratio/mean": 0.9777030008845031,
|
| 806 |
+
"sampling/importance_sampling_ratio/min": 0.7557966459542513,
|
| 807 |
+
"sampling/sampling_logp_difference/max": 0.5703878505155444,
|
| 808 |
+
"sampling/sampling_logp_difference/mean": 0.011606296713580377,
|
| 809 |
+
"step": 250,
|
| 810 |
+
"step_time": 10.26961963037029
|
| 811 |
+
}
|
| 812 |
+
],
|
| 813 |
+
"logging_steps": 10,
|
| 814 |
+
"max_steps": 256,
|
| 815 |
+
"num_input_tokens_seen": 5758627,
|
| 816 |
+
"num_train_epochs": 1,
|
| 817 |
+
"save_steps": 257,
|
| 818 |
+
"stateful_callbacks": {
|
| 819 |
+
"TrainerControl": {
|
| 820 |
+
"args": {
|
| 821 |
+
"should_epoch_stop": false,
|
| 822 |
+
"should_evaluate": false,
|
| 823 |
+
"should_log": false,
|
| 824 |
+
"should_save": true,
|
| 825 |
+
"should_training_stop": true
|
| 826 |
+
},
|
| 827 |
+
"attributes": {}
|
| 828 |
+
}
|
| 829 |
+
},
|
| 830 |
+
"total_flos": 0.0,
|
| 831 |
+
"train_batch_size": 4,
|
| 832 |
+
"trial_name": null,
|
| 833 |
+
"trial_params": null
|
| 834 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-32/adapter_config.json
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/workspace/rl3_direct_b/parent",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"layer_replication": null,
|
| 16 |
+
"layers_pattern": null,
|
| 17 |
+
"layers_to_transform": null,
|
| 18 |
+
"loftq_config": {},
|
| 19 |
+
"lora_alpha": 64,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 32,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"language_model.layers.13.self_attn.q_proj",
|
| 35 |
+
"46.self_attn.q_proj",
|
| 36 |
+
"39.self_attn.v_proj",
|
| 37 |
+
"language_model.layers.15.self_attn.q_proj",
|
| 38 |
+
"language_model.layers.19.self_attn.k_proj",
|
| 39 |
+
"language_model.layers.21.self_attn.v_proj",
|
| 40 |
+
"language_model.layers.9.self_attn.q_proj",
|
| 41 |
+
"41.self_attn.q_proj",
|
| 42 |
+
"o_proj",
|
| 43 |
+
"language_model.layers.16.self_attn.q_proj",
|
| 44 |
+
"language_model.layers.13.self_attn.k_proj",
|
| 45 |
+
"40.self_attn.k_proj",
|
| 46 |
+
"42.self_attn.k_proj",
|
| 47 |
+
"language_model.layers.18.self_attn.q_proj",
|
| 48 |
+
"language_model.layers.12.self_attn.k_proj",
|
| 49 |
+
"34.self_attn.k_proj",
|
| 50 |
+
"language_model.layers.0.self_attn.q_proj",
|
| 51 |
+
"34.self_attn.q_proj",
|
| 52 |
+
"37.self_attn.v_proj",
|
| 53 |
+
"language_model.layers.12.self_attn.q_proj",
|
| 54 |
+
"33.self_attn.k_proj",
|
| 55 |
+
"language_model.layers.8.self_attn.v_proj",
|
| 56 |
+
"32.self_attn.q_proj",
|
| 57 |
+
"43.self_attn.v_proj",
|
| 58 |
+
"language_model.layers.6.self_attn.q_proj",
|
| 59 |
+
"language_model.layers.7.self_attn.k_proj",
|
| 60 |
+
"language_model.layers.19.self_attn.q_proj",
|
| 61 |
+
"32.self_attn.k_proj",
|
| 62 |
+
"language_model.layers.23.self_attn.q_proj",
|
| 63 |
+
"language_model.layers.3.self_attn.q_proj",
|
| 64 |
+
"language_model.layers.22.self_attn.v_proj",
|
| 65 |
+
"36.self_attn.q_proj",
|
| 66 |
+
"33.self_attn.v_proj",
|
| 67 |
+
"language_model.layers.16.self_attn.k_proj",
|
| 68 |
+
"language_model.layers.1.self_attn.k_proj",
|
| 69 |
+
"language_model.layers.8.self_attn.k_proj",
|
| 70 |
+
"language_model.layers.15.self_attn.k_proj",
|
| 71 |
+
"33.self_attn.q_proj",
|
| 72 |
+
"language_model.layers.15.self_attn.v_proj",
|
| 73 |
+
"language_model.layers.26.self_attn.v_proj",
|
| 74 |
+
"42.self_attn.v_proj",
|
| 75 |
+
"29.self_attn.q_proj",
|
| 76 |
+
"45.self_attn.v_proj",
|
| 77 |
+
"language_model.layers.17.self_attn.q_proj",
|
| 78 |
+
"language_model.layers.24.self_attn.k_proj",
|
| 79 |
+
"language_model.layers.1.self_attn.v_proj",
|
| 80 |
+
"language_model.layers.20.self_attn.k_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"language_model.layers.5.self_attn.q_proj",
|
| 83 |
+
"34.self_attn.v_proj",
|
| 84 |
+
"44.self_attn.v_proj",
|
| 85 |
+
"language_model.layers.0.self_attn.k_proj",
|
| 86 |
+
"28.self_attn.v_proj",
|
| 87 |
+
"language_model.layers.22.self_attn.q_proj",
|
| 88 |
+
"30.self_attn.k_proj",
|
| 89 |
+
"30.self_attn.q_proj",
|
| 90 |
+
"language_model.layers.6.self_attn.v_proj",
|
| 91 |
+
"language_model.layers.14.self_attn.k_proj",
|
| 92 |
+
"language_model.layers.5.self_attn.v_proj",
|
| 93 |
+
"41.self_attn.v_proj",
|
| 94 |
+
"language_model.layers.25.self_attn.k_proj",
|
| 95 |
+
"31.self_attn.k_proj",
|
| 96 |
+
"language_model.layers.5.self_attn.k_proj",
|
| 97 |
+
"language_model.layers.4.self_attn.k_proj",
|
| 98 |
+
"35.self_attn.k_proj",
|
| 99 |
+
"37.self_attn.k_proj",
|
| 100 |
+
"language_model.layers.11.self_attn.k_proj",
|
| 101 |
+
"language_model.layers.8.self_attn.q_proj",
|
| 102 |
+
"29.self_attn.k_proj",
|
| 103 |
+
"47.self_attn.q_proj",
|
| 104 |
+
"language_model.layers.13.self_attn.v_proj",
|
| 105 |
+
"language_model.layers.9.self_attn.v_proj",
|
| 106 |
+
"language_model.layers.10.self_attn.q_proj",
|
| 107 |
+
"40.self_attn.v_proj",
|
| 108 |
+
"language_model.layers.10.self_attn.v_proj",
|
| 109 |
+
"up_proj",
|
| 110 |
+
"38.self_attn.v_proj",
|
| 111 |
+
"35.self_attn.q_proj",
|
| 112 |
+
"28.self_attn.k_proj",
|
| 113 |
+
"language_model.layers.18.self_attn.k_proj",
|
| 114 |
+
"language_model.layers.26.self_attn.q_proj",
|
| 115 |
+
"38.self_attn.q_proj",
|
| 116 |
+
"36.self_attn.v_proj",
|
| 117 |
+
"language_model.layers.11.self_attn.q_proj",
|
| 118 |
+
"language_model.layers.3.self_attn.k_proj",
|
| 119 |
+
"language_model.layers.24.self_attn.q_proj",
|
| 120 |
+
"37.self_attn.q_proj",
|
| 121 |
+
"45.self_attn.q_proj",
|
| 122 |
+
"42.self_attn.q_proj",
|
| 123 |
+
"language_model.layers.3.self_attn.v_proj",
|
| 124 |
+
"language_model.layers.0.self_attn.v_proj",
|
| 125 |
+
"language_model.layers.23.self_attn.k_proj",
|
| 126 |
+
"language_model.layers.9.self_attn.k_proj",
|
| 127 |
+
"27.self_attn.q_proj",
|
| 128 |
+
"language_model.layers.7.self_attn.q_proj",
|
| 129 |
+
"language_model.layers.25.self_attn.v_proj",
|
| 130 |
+
"30.self_attn.v_proj",
|
| 131 |
+
"language_model.layers.24.self_attn.v_proj",
|
| 132 |
+
"language_model.layers.14.self_attn.v_proj",
|
| 133 |
+
"language_model.layers.21.self_attn.q_proj",
|
| 134 |
+
"46.self_attn.v_proj",
|
| 135 |
+
"language_model.layers.20.self_attn.q_proj",
|
| 136 |
+
"31.self_attn.v_proj",
|
| 137 |
+
"28.self_attn.q_proj",
|
| 138 |
+
"27.self_attn.k_proj",
|
| 139 |
+
"40.self_attn.q_proj",
|
| 140 |
+
"language_model.layers.6.self_attn.k_proj",
|
| 141 |
+
"45.self_attn.k_proj",
|
| 142 |
+
"language_model.layers.1.self_attn.q_proj",
|
| 143 |
+
"38.self_attn.k_proj",
|
| 144 |
+
"language_model.layers.10.self_attn.k_proj",
|
| 145 |
+
"39.self_attn.k_proj",
|
| 146 |
+
"41.self_attn.k_proj",
|
| 147 |
+
"44.self_attn.k_proj",
|
| 148 |
+
"43.self_attn.q_proj",
|
| 149 |
+
"language_model.layers.2.self_attn.k_proj",
|
| 150 |
+
"language_model.layers.22.self_attn.k_proj",
|
| 151 |
+
"language_model.layers.18.self_attn.v_proj",
|
| 152 |
+
"language_model.layers.11.self_attn.v_proj",
|
| 153 |
+
"44.self_attn.q_proj",
|
| 154 |
+
"language_model.layers.2.self_attn.v_proj",
|
| 155 |
+
"down_proj",
|
| 156 |
+
"47.self_attn.v_proj",
|
| 157 |
+
"29.self_attn.v_proj",
|
| 158 |
+
"language_model.layers.21.self_attn.k_proj",
|
| 159 |
+
"35.self_attn.v_proj",
|
| 160 |
+
"46.self_attn.k_proj",
|
| 161 |
+
"language_model.layers.17.self_attn.k_proj",
|
| 162 |
+
"47.self_attn.k_proj",
|
| 163 |
+
"language_model.layers.7.self_attn.v_proj",
|
| 164 |
+
"39.self_attn.q_proj",
|
| 165 |
+
"language_model.layers.26.self_attn.k_proj",
|
| 166 |
+
"language_model.layers.2.self_attn.q_proj",
|
| 167 |
+
"language_model.layers.20.self_attn.v_proj",
|
| 168 |
+
"language_model.layers.14.self_attn.q_proj",
|
| 169 |
+
"language_model.layers.16.self_attn.v_proj",
|
| 170 |
+
"32.self_attn.v_proj",
|
| 171 |
+
"language_model.layers.23.self_attn.v_proj",
|
| 172 |
+
"language_model.layers.19.self_attn.v_proj",
|
| 173 |
+
"27.self_attn.v_proj",
|
| 174 |
+
"language_model.layers.4.self_attn.v_proj",
|
| 175 |
+
"language_model.layers.25.self_attn.q_proj",
|
| 176 |
+
"31.self_attn.q_proj",
|
| 177 |
+
"43.self_attn.k_proj",
|
| 178 |
+
"language_model.layers.4.self_attn.q_proj",
|
| 179 |
+
"language_model.layers.17.self_attn.v_proj",
|
| 180 |
+
"language_model.layers.12.self_attn.v_proj",
|
| 181 |
+
"36.self_attn.k_proj"
|
| 182 |
+
],
|
| 183 |
+
"target_parameters": null,
|
| 184 |
+
"task_type": "CAUSAL_LM",
|
| 185 |
+
"trainable_token_indices": null,
|
| 186 |
+
"use_bdlora": null,
|
| 187 |
+
"use_dora": false,
|
| 188 |
+
"use_qalora": false,
|
| 189 |
+
"use_rslora": false,
|
| 190 |
+
"velora_config": null
|
| 191 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-32/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:20f9d508cc8640337fd386363cd6c0f41035dc925829817f9657d632a29c4822
|
| 3 |
+
size 523864568
|
rl_grpo/coin_real_4x_direct/checkpoint-64/adapter_config.json
ADDED
|
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"alora_invocation_tokens": null,
|
| 3 |
+
"alpha_pattern": {},
|
| 4 |
+
"arrow_config": null,
|
| 5 |
+
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "/workspace/rl3_direct_b/parent",
|
| 7 |
+
"bias": "none",
|
| 8 |
+
"corda_config": null,
|
| 9 |
+
"ensure_weight_tying": false,
|
| 10 |
+
"eva_config": null,
|
| 11 |
+
"exclude_modules": null,
|
| 12 |
+
"fan_in_fan_out": false,
|
| 13 |
+
"inference_mode": true,
|
| 14 |
+
"init_lora_weights": true,
|
| 15 |
+
"layer_replication": null,
|
| 16 |
+
"layers_pattern": null,
|
| 17 |
+
"layers_to_transform": null,
|
| 18 |
+
"loftq_config": {},
|
| 19 |
+
"lora_alpha": 64,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 32,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"language_model.layers.13.self_attn.q_proj",
|
| 35 |
+
"46.self_attn.q_proj",
|
| 36 |
+
"39.self_attn.v_proj",
|
| 37 |
+
"language_model.layers.15.self_attn.q_proj",
|
| 38 |
+
"language_model.layers.19.self_attn.k_proj",
|
| 39 |
+
"language_model.layers.21.self_attn.v_proj",
|
| 40 |
+
"language_model.layers.9.self_attn.q_proj",
|
| 41 |
+
"41.self_attn.q_proj",
|
| 42 |
+
"o_proj",
|
| 43 |
+
"language_model.layers.16.self_attn.q_proj",
|
| 44 |
+
"language_model.layers.13.self_attn.k_proj",
|
| 45 |
+
"40.self_attn.k_proj",
|
| 46 |
+
"42.self_attn.k_proj",
|
| 47 |
+
"language_model.layers.18.self_attn.q_proj",
|
| 48 |
+
"language_model.layers.12.self_attn.k_proj",
|
| 49 |
+
"34.self_attn.k_proj",
|
| 50 |
+
"language_model.layers.0.self_attn.q_proj",
|
| 51 |
+
"34.self_attn.q_proj",
|
| 52 |
+
"37.self_attn.v_proj",
|
| 53 |
+
"language_model.layers.12.self_attn.q_proj",
|
| 54 |
+
"33.self_attn.k_proj",
|
| 55 |
+
"language_model.layers.8.self_attn.v_proj",
|
| 56 |
+
"32.self_attn.q_proj",
|
| 57 |
+
"43.self_attn.v_proj",
|
| 58 |
+
"language_model.layers.6.self_attn.q_proj",
|
| 59 |
+
"language_model.layers.7.self_attn.k_proj",
|
| 60 |
+
"language_model.layers.19.self_attn.q_proj",
|
| 61 |
+
"32.self_attn.k_proj",
|
| 62 |
+
"language_model.layers.23.self_attn.q_proj",
|
| 63 |
+
"language_model.layers.3.self_attn.q_proj",
|
| 64 |
+
"language_model.layers.22.self_attn.v_proj",
|
| 65 |
+
"36.self_attn.q_proj",
|
| 66 |
+
"33.self_attn.v_proj",
|
| 67 |
+
"language_model.layers.16.self_attn.k_proj",
|
| 68 |
+
"language_model.layers.1.self_attn.k_proj",
|
| 69 |
+
"language_model.layers.8.self_attn.k_proj",
|
| 70 |
+
"language_model.layers.15.self_attn.k_proj",
|
| 71 |
+
"33.self_attn.q_proj",
|
| 72 |
+
"language_model.layers.15.self_attn.v_proj",
|
| 73 |
+
"language_model.layers.26.self_attn.v_proj",
|
| 74 |
+
"42.self_attn.v_proj",
|
| 75 |
+
"29.self_attn.q_proj",
|
| 76 |
+
"45.self_attn.v_proj",
|
| 77 |
+
"language_model.layers.17.self_attn.q_proj",
|
| 78 |
+
"language_model.layers.24.self_attn.k_proj",
|
| 79 |
+
"language_model.layers.1.self_attn.v_proj",
|
| 80 |
+
"language_model.layers.20.self_attn.k_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"language_model.layers.5.self_attn.q_proj",
|
| 83 |
+
"34.self_attn.v_proj",
|
| 84 |
+
"44.self_attn.v_proj",
|
| 85 |
+
"language_model.layers.0.self_attn.k_proj",
|
| 86 |
+
"28.self_attn.v_proj",
|
| 87 |
+
"language_model.layers.22.self_attn.q_proj",
|
| 88 |
+
"30.self_attn.k_proj",
|
| 89 |
+
"30.self_attn.q_proj",
|
| 90 |
+
"language_model.layers.6.self_attn.v_proj",
|
| 91 |
+
"language_model.layers.14.self_attn.k_proj",
|
| 92 |
+
"language_model.layers.5.self_attn.v_proj",
|
| 93 |
+
"41.self_attn.v_proj",
|
| 94 |
+
"language_model.layers.25.self_attn.k_proj",
|
| 95 |
+
"31.self_attn.k_proj",
|
| 96 |
+
"language_model.layers.5.self_attn.k_proj",
|
| 97 |
+
"language_model.layers.4.self_attn.k_proj",
|
| 98 |
+
"35.self_attn.k_proj",
|
| 99 |
+
"37.self_attn.k_proj",
|
| 100 |
+
"language_model.layers.11.self_attn.k_proj",
|
| 101 |
+
"language_model.layers.8.self_attn.q_proj",
|
| 102 |
+
"29.self_attn.k_proj",
|
| 103 |
+
"47.self_attn.q_proj",
|
| 104 |
+
"language_model.layers.13.self_attn.v_proj",
|
| 105 |
+
"language_model.layers.9.self_attn.v_proj",
|
| 106 |
+
"language_model.layers.10.self_attn.q_proj",
|
| 107 |
+
"40.self_attn.v_proj",
|
| 108 |
+
"language_model.layers.10.self_attn.v_proj",
|
| 109 |
+
"up_proj",
|
| 110 |
+
"38.self_attn.v_proj",
|
| 111 |
+
"35.self_attn.q_proj",
|
| 112 |
+
"28.self_attn.k_proj",
|
| 113 |
+
"language_model.layers.18.self_attn.k_proj",
|
| 114 |
+
"language_model.layers.26.self_attn.q_proj",
|
| 115 |
+
"38.self_attn.q_proj",
|
| 116 |
+
"36.self_attn.v_proj",
|
| 117 |
+
"language_model.layers.11.self_attn.q_proj",
|
| 118 |
+
"language_model.layers.3.self_attn.k_proj",
|
| 119 |
+
"language_model.layers.24.self_attn.q_proj",
|
| 120 |
+
"37.self_attn.q_proj",
|
| 121 |
+
"45.self_attn.q_proj",
|
| 122 |
+
"42.self_attn.q_proj",
|
| 123 |
+
"language_model.layers.3.self_attn.v_proj",
|
| 124 |
+
"language_model.layers.0.self_attn.v_proj",
|
| 125 |
+
"language_model.layers.23.self_attn.k_proj",
|
| 126 |
+
"language_model.layers.9.self_attn.k_proj",
|
| 127 |
+
"27.self_attn.q_proj",
|
| 128 |
+
"language_model.layers.7.self_attn.q_proj",
|
| 129 |
+
"language_model.layers.25.self_attn.v_proj",
|
| 130 |
+
"30.self_attn.v_proj",
|
| 131 |
+
"language_model.layers.24.self_attn.v_proj",
|
| 132 |
+
"language_model.layers.14.self_attn.v_proj",
|
| 133 |
+
"language_model.layers.21.self_attn.q_proj",
|
| 134 |
+
"46.self_attn.v_proj",
|
| 135 |
+
"language_model.layers.20.self_attn.q_proj",
|
| 136 |
+
"31.self_attn.v_proj",
|
| 137 |
+
"28.self_attn.q_proj",
|
| 138 |
+
"27.self_attn.k_proj",
|
| 139 |
+
"40.self_attn.q_proj",
|
| 140 |
+
"language_model.layers.6.self_attn.k_proj",
|
| 141 |
+
"45.self_attn.k_proj",
|
| 142 |
+
"language_model.layers.1.self_attn.q_proj",
|
| 143 |
+
"38.self_attn.k_proj",
|
| 144 |
+
"language_model.layers.10.self_attn.k_proj",
|
| 145 |
+
"39.self_attn.k_proj",
|
| 146 |
+
"41.self_attn.k_proj",
|
| 147 |
+
"44.self_attn.k_proj",
|
| 148 |
+
"43.self_attn.q_proj",
|
| 149 |
+
"language_model.layers.2.self_attn.k_proj",
|
| 150 |
+
"language_model.layers.22.self_attn.k_proj",
|
| 151 |
+
"language_model.layers.18.self_attn.v_proj",
|
| 152 |
+
"language_model.layers.11.self_attn.v_proj",
|
| 153 |
+
"44.self_attn.q_proj",
|
| 154 |
+
"language_model.layers.2.self_attn.v_proj",
|
| 155 |
+
"down_proj",
|
| 156 |
+
"47.self_attn.v_proj",
|
| 157 |
+
"29.self_attn.v_proj",
|
| 158 |
+
"language_model.layers.21.self_attn.k_proj",
|
| 159 |
+
"35.self_attn.v_proj",
|
| 160 |
+
"46.self_attn.k_proj",
|
| 161 |
+
"language_model.layers.17.self_attn.k_proj",
|
| 162 |
+
"47.self_attn.k_proj",
|
| 163 |
+
"language_model.layers.7.self_attn.v_proj",
|
| 164 |
+
"39.self_attn.q_proj",
|
| 165 |
+
"language_model.layers.26.self_attn.k_proj",
|
| 166 |
+
"language_model.layers.2.self_attn.q_proj",
|
| 167 |
+
"language_model.layers.20.self_attn.v_proj",
|
| 168 |
+
"language_model.layers.14.self_attn.q_proj",
|
| 169 |
+
"language_model.layers.16.self_attn.v_proj",
|
| 170 |
+
"32.self_attn.v_proj",
|
| 171 |
+
"language_model.layers.23.self_attn.v_proj",
|
| 172 |
+
"language_model.layers.19.self_attn.v_proj",
|
| 173 |
+
"27.self_attn.v_proj",
|
| 174 |
+
"language_model.layers.4.self_attn.v_proj",
|
| 175 |
+
"language_model.layers.25.self_attn.q_proj",
|
| 176 |
+
"31.self_attn.q_proj",
|
| 177 |
+
"43.self_attn.k_proj",
|
| 178 |
+
"language_model.layers.4.self_attn.q_proj",
|
| 179 |
+
"language_model.layers.17.self_attn.v_proj",
|
| 180 |
+
"language_model.layers.12.self_attn.v_proj",
|
| 181 |
+
"36.self_attn.k_proj"
|
| 182 |
+
],
|
| 183 |
+
"target_parameters": null,
|
| 184 |
+
"task_type": "CAUSAL_LM",
|
| 185 |
+
"trainable_token_indices": null,
|
| 186 |
+
"use_bdlora": null,
|
| 187 |
+
"use_dora": false,
|
| 188 |
+
"use_qalora": false,
|
| 189 |
+
"use_rslora": false,
|
| 190 |
+
"velora_config": null
|
| 191 |
+
}
|
rl_grpo/coin_real_4x_direct/checkpoint-64/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:79033f10a28425c40084bd34d39c7625e799257924cd2572067fe933128a363e
|
| 3 |
+
size 523864568
|
rl_grpo/coin_real_4x_direct/checkpoint.json
ADDED
|
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"experiment": "scimt-train:dispatch-rl-coin_real_4x_direct",
|
| 3 |
+
"spec": null,
|
| 4 |
+
"kind": null,
|
| 5 |
+
"note": "Spec-free training stage (scimt.train.train_dataset) \u2014 a post-training link in a staged chain, not a spec install.",
|
| 6 |
+
"train": {
|
| 7 |
+
"data": "/workspace/rl3_direct_b/data/direct/train.jsonl",
|
| 8 |
+
"dataset_meta": {
|
| 9 |
+
"adhoc": true
|
| 10 |
+
},
|
| 11 |
+
"stage": null,
|
| 12 |
+
"seed": 42,
|
| 13 |
+
"load_checkpoint_path": "/workspace/rl3_direct_b/parent",
|
| 14 |
+
"grpo": {
|
| 15 |
+
"episodes": 8192,
|
| 16 |
+
"group_size": 8,
|
| 17 |
+
"max_prompt_length": 3072,
|
| 18 |
+
"max_completion_length": 256,
|
| 19 |
+
"per_device_batch_size": 4,
|
| 20 |
+
"gradient_accumulation_steps": 8,
|
| 21 |
+
"steps_per_generation": null,
|
| 22 |
+
"learning_rate": 1e-05,
|
| 23 |
+
"temperature": 0.7,
|
| 24 |
+
"loss_type": "dr_grpo",
|
| 25 |
+
"beta": 0.0,
|
| 26 |
+
"vllm": "colocate",
|
| 27 |
+
"vllm_gpu_memory_utilization": 0.45,
|
| 28 |
+
"vllm_max_model_len": 3328,
|
| 29 |
+
"mask_truncated_completions": true,
|
| 30 |
+
"log_completions": true,
|
| 31 |
+
"num_completions_to_print": 2,
|
| 32 |
+
"log_unique_prompts": true,
|
| 33 |
+
"checkpoint_fractions": [
|
| 34 |
+
0.0625,
|
| 35 |
+
0.125,
|
| 36 |
+
0.25,
|
| 37 |
+
0.5,
|
| 38 |
+
1.0
|
| 39 |
+
],
|
| 40 |
+
"report_to": [],
|
| 41 |
+
"reward_func": "experiments.prior_coins.dispatch_rl_reward_v2:reward_direct",
|
| 42 |
+
"resume_from_checkpoint": null,
|
| 43 |
+
"rollout_log_dir": "/workspace/rl3_direct_b/training/coin_real_4x_direct/logs",
|
| 44 |
+
"abort_log_path": null,
|
| 45 |
+
"validation_dataset_path": null,
|
| 46 |
+
"abort_eval_func": null,
|
| 47 |
+
"parent_agreement": null,
|
| 48 |
+
"parent_reward": null,
|
| 49 |
+
"parent_completion_length": null,
|
| 50 |
+
"zero_std_warmup_fraction": 0.1,
|
| 51 |
+
"completion_length_window": 1024
|
| 52 |
+
},
|
| 53 |
+
"lora": {
|
| 54 |
+
"r": 32,
|
| 55 |
+
"alpha": 64,
|
| 56 |
+
"dropout": 0.0,
|
| 57 |
+
"target_linear": true,
|
| 58 |
+
"target_modules": null
|
| 59 |
+
}
|
| 60 |
+
},
|
| 61 |
+
"run_name": "dispatch-rl-coin_real_4x_direct",
|
| 62 |
+
"pointer_file": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/ckpt_dispatch-rl-coin_real_4x_direct.txt",
|
| 63 |
+
"backend": "hf_grpo",
|
| 64 |
+
"sampler": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/sampler",
|
| 65 |
+
"state": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/trainer/checkpoint-256",
|
| 66 |
+
"model": "google/gemma-3-12b-pt",
|
| 67 |
+
"meta": {
|
| 68 |
+
"experiment": "scimt-train:dispatch-rl-coin_real_4x_direct",
|
| 69 |
+
"spec": null,
|
| 70 |
+
"kind": null,
|
| 71 |
+
"note": "Spec-free training stage (scimt.train.train_dataset) \u2014 a post-training link in a staged chain, not a spec install.",
|
| 72 |
+
"train": {
|
| 73 |
+
"data": "/workspace/rl3_direct_b/data/direct/train.jsonl",
|
| 74 |
+
"dataset_meta": {
|
| 75 |
+
"adhoc": true
|
| 76 |
+
},
|
| 77 |
+
"stage": null,
|
| 78 |
+
"seed": 42,
|
| 79 |
+
"load_checkpoint_path": "/workspace/rl3_direct_b/parent",
|
| 80 |
+
"grpo": {
|
| 81 |
+
"episodes": 8192,
|
| 82 |
+
"group_size": 8,
|
| 83 |
+
"max_prompt_length": 3072,
|
| 84 |
+
"max_completion_length": 256,
|
| 85 |
+
"per_device_batch_size": 4,
|
| 86 |
+
"gradient_accumulation_steps": 8,
|
| 87 |
+
"steps_per_generation": null,
|
| 88 |
+
"learning_rate": 1e-05,
|
| 89 |
+
"temperature": 0.7,
|
| 90 |
+
"loss_type": "dr_grpo",
|
| 91 |
+
"beta": 0.0,
|
| 92 |
+
"vllm": "colocate",
|
| 93 |
+
"vllm_gpu_memory_utilization": 0.45,
|
| 94 |
+
"vllm_max_model_len": 3328,
|
| 95 |
+
"mask_truncated_completions": true,
|
| 96 |
+
"log_completions": true,
|
| 97 |
+
"num_completions_to_print": 2,
|
| 98 |
+
"log_unique_prompts": true,
|
| 99 |
+
"checkpoint_fractions": [
|
| 100 |
+
0.0625,
|
| 101 |
+
0.125,
|
| 102 |
+
0.25,
|
| 103 |
+
0.5,
|
| 104 |
+
1.0
|
| 105 |
+
],
|
| 106 |
+
"report_to": [],
|
| 107 |
+
"reward_func": "experiments.prior_coins.dispatch_rl_reward_v2:reward_direct",
|
| 108 |
+
"resume_from_checkpoint": null,
|
| 109 |
+
"rollout_log_dir": "/workspace/rl3_direct_b/training/coin_real_4x_direct/logs",
|
| 110 |
+
"abort_log_path": null,
|
| 111 |
+
"validation_dataset_path": null,
|
| 112 |
+
"abort_eval_func": null,
|
| 113 |
+
"parent_agreement": null,
|
| 114 |
+
"parent_reward": null,
|
| 115 |
+
"parent_completion_length": null,
|
| 116 |
+
"zero_std_warmup_fraction": 0.1,
|
| 117 |
+
"completion_length_window": 1024
|
| 118 |
+
},
|
| 119 |
+
"lora": {
|
| 120 |
+
"r": 32,
|
| 121 |
+
"alpha": 64,
|
| 122 |
+
"dropout": 0.0,
|
| 123 |
+
"target_linear": true,
|
| 124 |
+
"target_modules": null
|
| 125 |
+
}
|
| 126 |
+
},
|
| 127 |
+
"run_name": "dispatch-rl-coin_real_4x_direct",
|
| 128 |
+
"pointer_file": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/ckpt_dispatch-rl-coin_real_4x_direct.txt"
|
| 129 |
+
},
|
| 130 |
+
"sampler_path": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/sampler",
|
| 131 |
+
"state_path": "/workspace/rl3_direct_b/training/coin_real_4x_direct/train/trainer/checkpoint-256"
|
| 132 |
+
}
|
rl_grpo/coin_real_4x_direct/logs/raw_rollouts.rank-0.jsonl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a9b06aac4d765ef305c2dde486b56740bd3795b206a6747e6a82cfb39a9708c6
|
| 3 |
+
size 169178635
|