Hugging Face's logo

radna
/
DeepSeek-R1-14B-RL-70-RLS3-ckpts

Safetensors
Model card Files
xet
Community
radna commited on
Commit
329f9c2
·
verified ·
1 Parent(s): d1d788c

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. checkpoint-10/adapter_config.json +2 -2
  2. checkpoint-10/trainer_state.json +15 -16
  3. checkpoint-10/training_args.bin +1 -1
  4. checkpoint-12/adapter_config.json +2 -2
  5. checkpoint-12/trainer_state.json +18 -19
  6. checkpoint-12/training_args.bin +1 -1
  7. checkpoint-14/adapter_config.json +2 -2
  8. checkpoint-14/trainer_state.json +20 -21
  9. checkpoint-14/training_args.bin +1 -1
  10. checkpoint-16/adapter_config.json +2 -2
  11. checkpoint-16/trainer_state.json +22 -23
  12. checkpoint-16/training_args.bin +1 -1
  13. checkpoint-18/adapter_config.json +2 -2
  14. checkpoint-18/trainer_state.json +25 -26
  15. checkpoint-18/training_args.bin +1 -1
  16. checkpoint-2/adapter_config.json +2 -2
  17. checkpoint-2/trainer_state.json +5 -6
  18. checkpoint-2/training_args.bin +1 -1
  19. checkpoint-20/adapter_config.json +2 -2
  20. checkpoint-20/trainer_state.json +27 -28
  21. checkpoint-20/training_args.bin +1 -1
  22. checkpoint-22/adapter_config.json +2 -2
  23. checkpoint-22/trainer_state.json +30 -31
  24. checkpoint-22/training_args.bin +1 -1
  25. checkpoint-24/adapter_config.json +2 -2
  26. checkpoint-24/trainer_state.json +33 -34
  27. checkpoint-24/training_args.bin +1 -1
  28. checkpoint-26/adapter_config.json +2 -2
  29. checkpoint-26/trainer_state.json +35 -36
  30. checkpoint-26/training_args.bin +1 -1
  31. checkpoint-28/adapter_config.json +2 -2
  32. checkpoint-28/trainer_state.json +37 -38
  33. checkpoint-28/training_args.bin +1 -1
  34. checkpoint-30/adapter_config.json +2 -2
  35. checkpoint-30/args.json +416 -0
  36. checkpoint-30/global_step30/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
  37. checkpoint-30/global_step30/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt +3 -0
  38. checkpoint-30/global_step30/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt +3 -0
  39. checkpoint-30/global_step30/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt +3 -0
  40. checkpoint-30/global_step30/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt +3 -0
  41. checkpoint-30/global_step30/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt +3 -0
  42. checkpoint-30/global_step30/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt +3 -0
  43. checkpoint-30/global_step30/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt +3 -0
  44. checkpoint-30/rng_state_0.pth +3 -0
  45. checkpoint-30/rng_state_1.pth +3 -0
  46. checkpoint-30/rng_state_2.pth +3 -0
  47. checkpoint-30/rng_state_3.pth +3 -0
  48. checkpoint-30/rng_state_4.pth +3 -0
  49. checkpoint-30/rng_state_5.pth +3 -0
  50. checkpoint-30/rng_state_6.pth +3 -0
checkpoint-10/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-10/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 6,
3
  "best_metric": 0.012996690347790718,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-6",
5
  "epoch": 2.4210526315789473,
@@ -7,7 +6,7 @@
7
  "global_step": 10,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  }
 
1
  {
 
2
  "best_metric": 0.012996690347790718,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-6",
4
  "epoch": 2.4210526315789473,
 
6
  "global_step": 10,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  }
checkpoint-10/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-12/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-12/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 12,
3
  "best_metric": 0.03234308212995529,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
5
  "epoch": 2.8421052631578947,
@@ -7,7 +6,7 @@
7
  "global_step": 12,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
 
1
  {
 
2
  "best_metric": 0.03234308212995529,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
4
  "epoch": 2.8421052631578947,
 
6
  "global_step": 12,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
checkpoint-12/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-14/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-14/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 12,
3
  "best_metric": 0.03234308212995529,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
5
  "epoch": 3.4210526315789473,
@@ -7,7 +6,7 @@
7
  "global_step": 14,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  }
 
1
  {
 
2
  "best_metric": 0.03234308212995529,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
4
  "epoch": 3.4210526315789473,
 
6
  "global_step": 14,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  }
checkpoint-14/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-16/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-16/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 12,
3
  "best_metric": 0.03234308212995529,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
5
  "epoch": 3.8421052631578947,
@@ -7,7 +6,7 @@
7
  "global_step": 16,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  }
 
1
  {
 
2
  "best_metric": 0.03234308212995529,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-12",
4
  "epoch": 3.8421052631578947,
 
6
  "global_step": 16,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  }
checkpoint-16/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-18/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-18/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 18,
3
  "best_metric": 0.03729328140616417,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
5
  "epoch": 4.421052631578947,
@@ -7,7 +6,7 @@
7
  "global_step": 18,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
 
1
  {
 
2
  "best_metric": 0.03729328140616417,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
4
  "epoch": 4.421052631578947,
 
6
  "global_step": 18,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
checkpoint-18/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-2/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-2/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": null,
3
  "best_metric": null,
4
  "best_model_checkpoint": null,
5
  "epoch": 0.42105263157894735,
@@ -7,7 +6,7 @@
7
  "global_step": 2,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  }
40
  ],
41
  "logging_steps": 1,
 
1
  {
 
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
  "epoch": 0.42105263157894735,
 
6
  "global_step": 2,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  }
39
  ],
40
  "logging_steps": 1,
checkpoint-2/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-20/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-20/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 18,
3
  "best_metric": 0.03729328140616417,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
5
  "epoch": 4.842105263157895,
@@ -7,7 +6,7 @@
7
  "global_step": 20,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
@@ -311,7 +310,7 @@
311
  "kl": 0.1820068359375,
312
  "learning_rate": 8.636868207865244e-05,
313
  "loss": -0.03466903418302536,
314
- "memory(GiB)": 187.02,
315
  "response_clip_ratio": 0.466796875,
316
  "reward": 0.04069916973821819,
317
  "reward_std": 0.11991005763411522,
@@ -327,7 +326,7 @@
327
  "kl": 0.19287109375,
328
  "learning_rate": 8.43120818934367e-05,
329
  "loss": -0.03502114117145538,
330
- "memory(GiB)": 187.02,
331
  "step": 20,
332
  "train_speed(iter/s)": 0.000424
333
  }
 
1
  {
 
2
  "best_metric": 0.03729328140616417,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
4
  "epoch": 4.842105263157895,
 
6
  "global_step": 20,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
 
310
  "kl": 0.1820068359375,
311
  "learning_rate": 8.636868207865244e-05,
312
  "loss": -0.03466903418302536,
313
+ "memory(GiB)": 182.91,
314
  "response_clip_ratio": 0.466796875,
315
  "reward": 0.04069916973821819,
316
  "reward_std": 0.11991005763411522,
 
326
  "kl": 0.19287109375,
327
  "learning_rate": 8.43120818934367e-05,
328
  "loss": -0.03502114117145538,
329
+ "memory(GiB)": 182.91,
330
  "step": 20,
331
  "train_speed(iter/s)": 0.000424
332
  }
checkpoint-20/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-22/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-22/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 18,
3
  "best_metric": 0.03729328140616417,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
5
  "epoch": 5.421052631578947,
@@ -7,7 +6,7 @@
7
  "global_step": 22,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
@@ -311,7 +310,7 @@
311
  "kl": 0.1820068359375,
312
  "learning_rate": 8.636868207865244e-05,
313
  "loss": -0.03466903418302536,
314
- "memory(GiB)": 187.02,
315
  "response_clip_ratio": 0.466796875,
316
  "reward": 0.04069916973821819,
317
  "reward_std": 0.11991005763411522,
@@ -327,7 +326,7 @@
327
  "kl": 0.19287109375,
328
  "learning_rate": 8.43120818934367e-05,
329
  "loss": -0.03502114117145538,
330
- "memory(GiB)": 187.02,
331
  "step": 20,
332
  "train_speed(iter/s)": 0.000424
333
  },
@@ -339,14 +338,14 @@
339
  "kl": 0.17626953125,
340
  "learning_rate": 8.213938048432697e-05,
341
  "loss": -0.008662773296236992,
342
- "memory(GiB)": 187.02,
343
  "response_clip_ratio": 0.5625,
344
  "reward": 0.04996980866417289,
345
  "reward_std": 0.13849420100450516,
346
  "rewards/CosineReward": 0.049969930201768875,
347
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
348
  "step": 21,
349
- "train_speed(iter/s)": 0.000407
350
  },
351
  {
352
  "clip_ratio": 5.869188044016482e-05,
@@ -355,7 +354,7 @@
355
  "kl": 0.178955078125,
356
  "learning_rate": 7.985792958513931e-05,
357
  "loss": -0.008743642829358578,
358
- "memory(GiB)": 187.02,
359
  "step": 22,
360
  "train_speed(iter/s)": 0.000426
361
  }
 
1
  {
 
2
  "best_metric": 0.03729328140616417,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-18",
4
  "epoch": 5.421052631578947,
 
6
  "global_step": 22,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
 
310
  "kl": 0.1820068359375,
311
  "learning_rate": 8.636868207865244e-05,
312
  "loss": -0.03466903418302536,
313
+ "memory(GiB)": 182.91,
314
  "response_clip_ratio": 0.466796875,
315
  "reward": 0.04069916973821819,
316
  "reward_std": 0.11991005763411522,
 
326
  "kl": 0.19287109375,
327
  "learning_rate": 8.43120818934367e-05,
328
  "loss": -0.03502114117145538,
329
+ "memory(GiB)": 182.91,
330
  "step": 20,
331
  "train_speed(iter/s)": 0.000424
332
  },
 
338
  "kl": 0.17626953125,
339
  "learning_rate": 8.213938048432697e-05,
340
  "loss": -0.008662773296236992,
341
+ "memory(GiB)": 182.91,
342
  "response_clip_ratio": 0.5625,
343
  "reward": 0.04996980866417289,
344
  "reward_std": 0.13849420100450516,
345
  "rewards/CosineReward": 0.049969930201768875,
346
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
347
  "step": 21,
348
+ "train_speed(iter/s)": 0.000408
349
  },
350
  {
351
  "clip_ratio": 5.869188044016482e-05,
 
354
  "kl": 0.178955078125,
355
  "learning_rate": 7.985792958513931e-05,
356
  "loss": -0.008743642829358578,
357
+ "memory(GiB)": 182.91,
358
  "step": 22,
359
  "train_speed(iter/s)": 0.000426
360
  }
checkpoint-22/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-24/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-24/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 24,
3
  "best_metric": 0.04339282959699631,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
5
  "epoch": 5.842105263157895,
@@ -7,7 +6,7 @@
7
  "global_step": 24,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
@@ -311,7 +310,7 @@
311
  "kl": 0.1820068359375,
312
  "learning_rate": 8.636868207865244e-05,
313
  "loss": -0.03466903418302536,
314
- "memory(GiB)": 187.02,
315
  "response_clip_ratio": 0.466796875,
316
  "reward": 0.04069916973821819,
317
  "reward_std": 0.11991005763411522,
@@ -327,7 +326,7 @@
327
  "kl": 0.19287109375,
328
  "learning_rate": 8.43120818934367e-05,
329
  "loss": -0.03502114117145538,
330
- "memory(GiB)": 187.02,
331
  "step": 20,
332
  "train_speed(iter/s)": 0.000424
333
  },
@@ -339,14 +338,14 @@
339
  "kl": 0.17626953125,
340
  "learning_rate": 8.213938048432697e-05,
341
  "loss": -0.008662773296236992,
342
- "memory(GiB)": 187.02,
343
  "response_clip_ratio": 0.5625,
344
  "reward": 0.04996980866417289,
345
  "reward_std": 0.13849420100450516,
346
  "rewards/CosineReward": 0.049969930201768875,
347
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
348
  "step": 21,
349
- "train_speed(iter/s)": 0.000407
350
  },
351
  {
352
  "clip_ratio": 5.869188044016482e-05,
@@ -355,7 +354,7 @@
355
  "kl": 0.178955078125,
356
  "learning_rate": 7.985792958513931e-05,
357
  "loss": -0.008743642829358578,
358
- "memory(GiB)": 187.02,
359
  "step": 22,
360
  "train_speed(iter/s)": 0.000426
361
  },
@@ -367,7 +366,7 @@
367
  "kl": 0.1796875,
368
  "learning_rate": 7.74754489035403e-05,
369
  "loss": -0.03423420712351799,
370
- "memory(GiB)": 187.02,
371
  "response_clip_ratio": 0.583984375,
372
  "reward": 0.034468831261619925,
373
  "reward_std": 0.11841745302081108,
@@ -381,7 +380,7 @@
381
  "grad_norm": 0.014131724834442139,
382
  "learning_rate": 7.500000000000001e-05,
383
  "loss": -0.03426633030176163,
384
- "memory(GiB)": 187.02,
385
  "step": 24,
386
  "train_speed(iter/s)": 0.000427
387
  },
@@ -396,7 +395,7 @@
396
  "eval_reward_std": 0.10456253588199615,
397
  "eval_rewards/CosineReward": 0.04339282959699631,
398
  "eval_rewards/RepetitionPenalty": 0.0,
399
- "eval_runtime": 1045.0642,
400
  "eval_samples_per_second": 0.001,
401
  "eval_steps_per_second": 0.001,
402
  "step": 24
 
1
  {
 
2
  "best_metric": 0.04339282959699631,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
4
  "epoch": 5.842105263157895,
 
6
  "global_step": 24,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
 
310
  "kl": 0.1820068359375,
311
  "learning_rate": 8.636868207865244e-05,
312
  "loss": -0.03466903418302536,
313
+ "memory(GiB)": 182.91,
314
  "response_clip_ratio": 0.466796875,
315
  "reward": 0.04069916973821819,
316
  "reward_std": 0.11991005763411522,
 
326
  "kl": 0.19287109375,
327
  "learning_rate": 8.43120818934367e-05,
328
  "loss": -0.03502114117145538,
329
+ "memory(GiB)": 182.91,
330
  "step": 20,
331
  "train_speed(iter/s)": 0.000424
332
  },
 
338
  "kl": 0.17626953125,
339
  "learning_rate": 8.213938048432697e-05,
340
  "loss": -0.008662773296236992,
341
+ "memory(GiB)": 182.91,
342
  "response_clip_ratio": 0.5625,
343
  "reward": 0.04996980866417289,
344
  "reward_std": 0.13849420100450516,
345
  "rewards/CosineReward": 0.049969930201768875,
346
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
347
  "step": 21,
348
+ "train_speed(iter/s)": 0.000408
349
  },
350
  {
351
  "clip_ratio": 5.869188044016482e-05,
 
354
  "kl": 0.178955078125,
355
  "learning_rate": 7.985792958513931e-05,
356
  "loss": -0.008743642829358578,
357
+ "memory(GiB)": 182.91,
358
  "step": 22,
359
  "train_speed(iter/s)": 0.000426
360
  },
 
366
  "kl": 0.1796875,
367
  "learning_rate": 7.74754489035403e-05,
368
  "loss": -0.03423420712351799,
369
+ "memory(GiB)": 182.91,
370
  "response_clip_ratio": 0.583984375,
371
  "reward": 0.034468831261619925,
372
  "reward_std": 0.11841745302081108,
 
380
  "grad_norm": 0.014131724834442139,
381
  "learning_rate": 7.500000000000001e-05,
382
  "loss": -0.03426633030176163,
383
+ "memory(GiB)": 182.91,
384
  "step": 24,
385
  "train_speed(iter/s)": 0.000427
386
  },
 
395
  "eval_reward_std": 0.10456253588199615,
396
  "eval_rewards/CosineReward": 0.04339282959699631,
397
  "eval_rewards/RepetitionPenalty": 0.0,
398
+ "eval_runtime": 1045.0632,
399
  "eval_samples_per_second": 0.001,
400
  "eval_steps_per_second": 0.001,
401
  "step": 24
checkpoint-24/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-26/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-26/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 24,
3
  "best_metric": 0.04339282959699631,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
5
  "epoch": 6.421052631578947,
@@ -7,7 +6,7 @@
7
  "global_step": 26,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
@@ -311,7 +310,7 @@
311
  "kl": 0.1820068359375,
312
  "learning_rate": 8.636868207865244e-05,
313
  "loss": -0.03466903418302536,
314
- "memory(GiB)": 187.02,
315
  "response_clip_ratio": 0.466796875,
316
  "reward": 0.04069916973821819,
317
  "reward_std": 0.11991005763411522,
@@ -327,7 +326,7 @@
327
  "kl": 0.19287109375,
328
  "learning_rate": 8.43120818934367e-05,
329
  "loss": -0.03502114117145538,
330
- "memory(GiB)": 187.02,
331
  "step": 20,
332
  "train_speed(iter/s)": 0.000424
333
  },
@@ -339,14 +338,14 @@
339
  "kl": 0.17626953125,
340
  "learning_rate": 8.213938048432697e-05,
341
  "loss": -0.008662773296236992,
342
- "memory(GiB)": 187.02,
343
  "response_clip_ratio": 0.5625,
344
  "reward": 0.04996980866417289,
345
  "reward_std": 0.13849420100450516,
346
  "rewards/CosineReward": 0.049969930201768875,
347
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
348
  "step": 21,
349
- "train_speed(iter/s)": 0.000407
350
  },
351
  {
352
  "clip_ratio": 5.869188044016482e-05,
@@ -355,7 +354,7 @@
355
  "kl": 0.178955078125,
356
  "learning_rate": 7.985792958513931e-05,
357
  "loss": -0.008743642829358578,
358
- "memory(GiB)": 187.02,
359
  "step": 22,
360
  "train_speed(iter/s)": 0.000426
361
  },
@@ -367,7 +366,7 @@
367
  "kl": 0.1796875,
368
  "learning_rate": 7.74754489035403e-05,
369
  "loss": -0.03423420712351799,
370
- "memory(GiB)": 187.02,
371
  "response_clip_ratio": 0.583984375,
372
  "reward": 0.034468831261619925,
373
  "reward_std": 0.11841745302081108,
@@ -381,7 +380,7 @@
381
  "grad_norm": 0.014131724834442139,
382
  "learning_rate": 7.500000000000001e-05,
383
  "loss": -0.03426633030176163,
384
- "memory(GiB)": 187.02,
385
  "step": 24,
386
  "train_speed(iter/s)": 0.000427
387
  },
@@ -396,7 +395,7 @@
396
  "eval_reward_std": 0.10456253588199615,
397
  "eval_rewards/CosineReward": 0.04339282959699631,
398
  "eval_rewards/RepetitionPenalty": 0.0,
399
- "eval_runtime": 1045.0642,
400
  "eval_samples_per_second": 0.001,
401
  "eval_steps_per_second": 0.001,
402
  "step": 24
@@ -409,7 +408,7 @@
409
  "kl": 0.1800537109375,
410
  "learning_rate": 7.243995901002312e-05,
411
  "loss": -0.02097315341234207,
412
- "memory(GiB)": 187.02,
413
  "response_clip_ratio": 0.6171875,
414
  "reward": 0.03010205877944827,
415
  "reward_std": 0.10742511600255966,
@@ -425,7 +424,7 @@
425
  "kl": 0.18408203125,
426
  "learning_rate": 6.980398830195785e-05,
427
  "loss": -0.02103913575410843,
428
- "memory(GiB)": 187.02,
429
  "step": 26,
430
  "train_speed(iter/s)": 0.000421
431
  }
 
1
  {
 
2
  "best_metric": 0.04339282959699631,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
4
  "epoch": 6.421052631578947,
 
6
  "global_step": 26,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
 
310
  "kl": 0.1820068359375,
311
  "learning_rate": 8.636868207865244e-05,
312
  "loss": -0.03466903418302536,
313
+ "memory(GiB)": 182.91,
314
  "response_clip_ratio": 0.466796875,
315
  "reward": 0.04069916973821819,
316
  "reward_std": 0.11991005763411522,
 
326
  "kl": 0.19287109375,
327
  "learning_rate": 8.43120818934367e-05,
328
  "loss": -0.03502114117145538,
329
+ "memory(GiB)": 182.91,
330
  "step": 20,
331
  "train_speed(iter/s)": 0.000424
332
  },
 
338
  "kl": 0.17626953125,
339
  "learning_rate": 8.213938048432697e-05,
340
  "loss": -0.008662773296236992,
341
+ "memory(GiB)": 182.91,
342
  "response_clip_ratio": 0.5625,
343
  "reward": 0.04996980866417289,
344
  "reward_std": 0.13849420100450516,
345
  "rewards/CosineReward": 0.049969930201768875,
346
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
347
  "step": 21,
348
+ "train_speed(iter/s)": 0.000408
349
  },
350
  {
351
  "clip_ratio": 5.869188044016482e-05,
 
354
  "kl": 0.178955078125,
355
  "learning_rate": 7.985792958513931e-05,
356
  "loss": -0.008743642829358578,
357
+ "memory(GiB)": 182.91,
358
  "step": 22,
359
  "train_speed(iter/s)": 0.000426
360
  },
 
366
  "kl": 0.1796875,
367
  "learning_rate": 7.74754489035403e-05,
368
  "loss": -0.03423420712351799,
369
+ "memory(GiB)": 182.91,
370
  "response_clip_ratio": 0.583984375,
371
  "reward": 0.034468831261619925,
372
  "reward_std": 0.11841745302081108,
 
380
  "grad_norm": 0.014131724834442139,
381
  "learning_rate": 7.500000000000001e-05,
382
  "loss": -0.03426633030176163,
383
+ "memory(GiB)": 182.91,
384
  "step": 24,
385
  "train_speed(iter/s)": 0.000427
386
  },
 
395
  "eval_reward_std": 0.10456253588199615,
396
  "eval_rewards/CosineReward": 0.04339282959699631,
397
  "eval_rewards/RepetitionPenalty": 0.0,
398
+ "eval_runtime": 1045.0632,
399
  "eval_samples_per_second": 0.001,
400
  "eval_steps_per_second": 0.001,
401
  "step": 24
 
408
  "kl": 0.1800537109375,
409
  "learning_rate": 7.243995901002312e-05,
410
  "loss": -0.02097315341234207,
411
+ "memory(GiB)": 182.91,
412
  "response_clip_ratio": 0.6171875,
413
  "reward": 0.03010205877944827,
414
  "reward_std": 0.10742511600255966,
 
424
  "kl": 0.18408203125,
425
  "learning_rate": 6.980398830195785e-05,
426
  "loss": -0.02103913575410843,
427
+ "memory(GiB)": 182.91,
428
  "step": 26,
429
  "train_speed(iter/s)": 0.000421
430
  }
checkpoint-26/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-28/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-28/trainer_state.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "best_global_step": 24,
3
  "best_metric": 0.04339282959699631,
4
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
5
  "epoch": 6.842105263157895,
@@ -7,7 +6,7 @@
7
  "global_step": 28,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
- "is_world_process_zero": false,
11
  "log_history": [
12
  {
13
  "clip_ratio": 0.0,
@@ -17,14 +16,14 @@
17
  "kl": 0.0,
18
  "learning_rate": 1.6666666666666667e-05,
19
  "loss": -0.11016345024108887,
20
- "memory(GiB)": 176.98,
21
  "response_clip_ratio": 0.11328125,
22
  "reward": -0.002658387296833098,
23
  "reward_std": 0.06134121119976044,
24
  "rewards/CosineReward": -0.0026579967816360295,
25
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
26
  "step": 1,
27
- "train_speed(iter/s)": 0.000241
28
  },
29
  {
30
  "clip_ratio": 0.0,
@@ -33,9 +32,9 @@
33
  "kl": 0.0,
34
  "learning_rate": 3.3333333333333335e-05,
35
  "loss": -0.11016345024108887,
36
- "memory(GiB)": 176.98,
37
  "step": 2,
38
- "train_speed(iter/s)": 0.000466
39
  },
40
  {
41
  "clip_ratio": 1.3441811461234465e-05,
@@ -45,7 +44,7 @@
45
  "kl": 9.50181856751442e-07,
46
  "learning_rate": 5e-05,
47
  "loss": -0.06604708731174469,
48
- "memory(GiB)": 176.98,
49
  "response_clip_ratio": 0.13671875,
50
  "reward": 0.0006296975770965219,
51
  "reward_std": 0.07172460854053497,
@@ -61,7 +60,7 @@
61
  "kl": 1.1101365089416504e-05,
62
  "learning_rate": 6.666666666666667e-05,
63
  "loss": -0.06727766245603561,
64
- "memory(GiB)": 176.98,
65
  "step": 4,
66
  "train_speed(iter/s)": 0.000458
67
  },
@@ -73,7 +72,7 @@
73
  "kl": 0.00017762184143066406,
74
  "learning_rate": 8.333333333333334e-05,
75
  "loss": -0.09315311908721924,
76
- "memory(GiB)": 176.98,
77
  "response_clip_ratio": 0.119140625,
78
  "reward": -0.005135859013535082,
79
  "reward_std": 0.07994875870645046,
@@ -87,9 +86,9 @@
87
  "grad_norm": 0.18263348937034607,
88
  "learning_rate": 0.0001,
89
  "loss": -0.1041698157787323,
90
- "memory(GiB)": 176.98,
91
  "step": 6,
92
- "train_speed(iter/s)": 0.000458
93
  },
94
  {
95
  "epoch": 1.4210526315789473,
@@ -102,7 +101,7 @@
102
  "eval_reward_std": 0.08769983053207397,
103
  "eval_rewards/CosineReward": 0.012996694073081017,
104
  "eval_rewards/RepetitionPenalty": 0.0,
105
- "eval_runtime": 1030.1126,
106
  "eval_samples_per_second": 0.001,
107
  "eval_steps_per_second": 0.001,
108
  "step": 6
@@ -115,7 +114,7 @@
115
  "kl": 0.017406463623046875,
116
  "learning_rate": 9.991540791356342e-05,
117
  "loss": -0.051375165581703186,
118
- "memory(GiB)": 176.98,
119
  "response_clip_ratio": 0.1484375,
120
  "reward": 0.004909618757665157,
121
  "reward_std": 0.08167182095348835,
@@ -131,7 +130,7 @@
131
  "kl": 0.089599609375,
132
  "learning_rate": 9.966191788709716e-05,
133
  "loss": -0.05105742812156677,
134
- "memory(GiB)": 176.98,
135
  "step": 8,
136
  "train_speed(iter/s)": 0.000433
137
  },
@@ -143,7 +142,7 @@
143
  "kl": 0.0963134765625,
144
  "learning_rate": 9.924038765061042e-05,
145
  "loss": -0.05842069163918495,
146
- "memory(GiB)": 176.98,
147
  "response_clip_ratio": 0.255859375,
148
  "reward": 0.03643610421568155,
149
  "reward_std": 0.11898956261575222,
@@ -159,7 +158,7 @@
159
  "kl": 0.1185302734375,
160
  "learning_rate": 9.865224352899119e-05,
161
  "loss": -0.06491819024085999,
162
- "memory(GiB)": 176.98,
163
  "step": 10,
164
  "train_speed(iter/s)": 0.000436
165
  },
@@ -171,7 +170,7 @@
171
  "kl": 0.1275634765625,
172
  "learning_rate": 9.789947561577445e-05,
173
  "loss": -0.04600231721997261,
174
- "memory(GiB)": 187.02,
175
  "response_clip_ratio": 0.361328125,
176
  "reward": 0.023204635945148766,
177
  "reward_std": 0.10593634657561779,
@@ -185,7 +184,7 @@
185
  "grad_norm": 0.05781339108943939,
186
  "learning_rate": 9.698463103929542e-05,
187
  "loss": -0.05069056898355484,
188
- "memory(GiB)": 187.02,
189
  "step": 12,
190
  "train_speed(iter/s)": 0.000439
191
  },
@@ -200,7 +199,7 @@
200
  "eval_reward_std": 0.10685288906097412,
201
  "eval_rewards/CosineReward": 0.03234308212995529,
202
  "eval_rewards/RepetitionPenalty": 0.0,
203
- "eval_runtime": 1025.9048,
204
  "eval_samples_per_second": 0.001,
205
  "eval_steps_per_second": 0.001,
206
  "step": 12
@@ -213,7 +212,7 @@
213
  "kl": 0.151123046875,
214
  "learning_rate": 9.591080534401371e-05,
215
  "loss": -0.02191038429737091,
216
- "memory(GiB)": 187.02,
217
  "response_clip_ratio": 0.419921875,
218
  "reward": 0.035983758978545666,
219
  "reward_std": 0.11553369648754597,
@@ -229,7 +228,7 @@
229
  "kl": 0.169189453125,
230
  "learning_rate": 9.468163201617062e-05,
231
  "loss": -0.022672578692436218,
232
- "memory(GiB)": 187.02,
233
  "step": 14,
234
  "train_speed(iter/s)": 0.000427
235
  },
@@ -241,7 +240,7 @@
241
  "kl": 0.166748046875,
242
  "learning_rate": 9.330127018922194e-05,
243
  "loss": -0.059799157083034515,
244
- "memory(GiB)": 187.02,
245
  "response_clip_ratio": 0.4765625,
246
  "reward": 0.03584331553429365,
247
  "reward_std": 0.11829411797225475,
@@ -257,7 +256,7 @@
257
  "kl": 0.16748046875,
258
  "learning_rate": 9.177439057064683e-05,
259
  "loss": -0.06071458384394646,
260
- "memory(GiB)": 187.02,
261
  "step": 16,
262
  "train_speed(iter/s)": 0.000431
263
  },
@@ -269,7 +268,7 @@
269
  "kl": 0.1787109375,
270
  "learning_rate": 9.01061596377522e-05,
271
  "loss": -0.04504441097378731,
272
- "memory(GiB)": 187.02,
273
  "response_clip_ratio": 0.5625,
274
  "reward": 0.027318883687257767,
275
  "reward_std": 0.10441224090754986,
@@ -283,7 +282,7 @@
283
  "grad_norm": 0.005998397711664438,
284
  "learning_rate": 8.83022221559489e-05,
285
  "loss": -0.045487549155950546,
286
- "memory(GiB)": 187.02,
287
  "step": 18,
288
  "train_speed(iter/s)": 0.000432
289
  },
@@ -298,7 +297,7 @@
298
  "eval_reward_std": 0.10691346973180771,
299
  "eval_rewards/CosineReward": 0.03729327768087387,
300
  "eval_rewards/RepetitionPenalty": 0.0,
301
- "eval_runtime": 1041.2321,
302
  "eval_samples_per_second": 0.001,
303
  "eval_steps_per_second": 0.001,
304
  "step": 18
@@ -311,7 +310,7 @@
311
  "kl": 0.1820068359375,
312
  "learning_rate": 8.636868207865244e-05,
313
  "loss": -0.03466903418302536,
314
- "memory(GiB)": 187.02,
315
  "response_clip_ratio": 0.466796875,
316
  "reward": 0.04069916973821819,
317
  "reward_std": 0.11991005763411522,
@@ -327,7 +326,7 @@
327
  "kl": 0.19287109375,
328
  "learning_rate": 8.43120818934367e-05,
329
  "loss": -0.03502114117145538,
330
- "memory(GiB)": 187.02,
331
  "step": 20,
332
  "train_speed(iter/s)": 0.000424
333
  },
@@ -339,14 +338,14 @@
339
  "kl": 0.17626953125,
340
  "learning_rate": 8.213938048432697e-05,
341
  "loss": -0.008662773296236992,
342
- "memory(GiB)": 187.02,
343
  "response_clip_ratio": 0.5625,
344
  "reward": 0.04996980866417289,
345
  "reward_std": 0.13849420100450516,
346
  "rewards/CosineReward": 0.049969930201768875,
347
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
348
  "step": 21,
349
- "train_speed(iter/s)": 0.000407
350
  },
351
  {
352
  "clip_ratio": 5.869188044016482e-05,
@@ -355,7 +354,7 @@
355
  "kl": 0.178955078125,
356
  "learning_rate": 7.985792958513931e-05,
357
  "loss": -0.008743642829358578,
358
- "memory(GiB)": 187.02,
359
  "step": 22,
360
  "train_speed(iter/s)": 0.000426
361
  },
@@ -367,7 +366,7 @@
367
  "kl": 0.1796875,
368
  "learning_rate": 7.74754489035403e-05,
369
  "loss": -0.03423420712351799,
370
- "memory(GiB)": 187.02,
371
  "response_clip_ratio": 0.583984375,
372
  "reward": 0.034468831261619925,
373
  "reward_std": 0.11841745302081108,
@@ -381,7 +380,7 @@
381
  "grad_norm": 0.014131724834442139,
382
  "learning_rate": 7.500000000000001e-05,
383
  "loss": -0.03426633030176163,
384
- "memory(GiB)": 187.02,
385
  "step": 24,
386
  "train_speed(iter/s)": 0.000427
387
  },
@@ -396,7 +395,7 @@
396
  "eval_reward_std": 0.10456253588199615,
397
  "eval_rewards/CosineReward": 0.04339282959699631,
398
  "eval_rewards/RepetitionPenalty": 0.0,
399
- "eval_runtime": 1045.0642,
400
  "eval_samples_per_second": 0.001,
401
  "eval_steps_per_second": 0.001,
402
  "step": 24
@@ -409,7 +408,7 @@
409
  "kl": 0.1800537109375,
410
  "learning_rate": 7.243995901002312e-05,
411
  "loss": -0.02097315341234207,
412
- "memory(GiB)": 187.02,
413
  "response_clip_ratio": 0.6171875,
414
  "reward": 0.03010205877944827,
415
  "reward_std": 0.10742511600255966,
@@ -425,7 +424,7 @@
425
  "kl": 0.18408203125,
426
  "learning_rate": 6.980398830195785e-05,
427
  "loss": -0.02103913575410843,
428
- "memory(GiB)": 187.02,
429
  "step": 26,
430
  "train_speed(iter/s)": 0.000421
431
  },
@@ -437,7 +436,7 @@
437
  "kl": 0.174560546875,
438
  "learning_rate": 6.710100716628344e-05,
439
  "loss": -0.03593946248292923,
440
- "memory(GiB)": 187.02,
441
  "response_clip_ratio": 0.513671875,
442
  "reward": 0.04752760287374258,
443
  "reward_std": 0.14935147762298584,
@@ -453,7 +452,7 @@
453
  "kl": 0.182373046875,
454
  "learning_rate": 6.434016163555452e-05,
455
  "loss": -0.03595500811934471,
456
- "memory(GiB)": 187.02,
457
  "step": 28,
458
  "train_speed(iter/s)": 0.000422
459
  }
 
1
  {
 
2
  "best_metric": 0.04339282959699631,
3
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
4
  "epoch": 6.842105263157895,
 
6
  "global_step": 28,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "clip_ratio": 0.0,
 
16
  "kl": 0.0,
17
  "learning_rate": 1.6666666666666667e-05,
18
  "loss": -0.11016345024108887,
19
+ "memory(GiB)": 182.91,
20
  "response_clip_ratio": 0.11328125,
21
  "reward": -0.002658387296833098,
22
  "reward_std": 0.06134121119976044,
23
  "rewards/CosineReward": -0.0026579967816360295,
24
  "rewards/RepetitionPenalty": -3.8975886695880035e-07,
25
  "step": 1,
26
+ "train_speed(iter/s)": 0.000242
27
  },
28
  {
29
  "clip_ratio": 0.0,
 
32
  "kl": 0.0,
33
  "learning_rate": 3.3333333333333335e-05,
34
  "loss": -0.11016345024108887,
35
+ "memory(GiB)": 182.91,
36
  "step": 2,
37
+ "train_speed(iter/s)": 0.000467
38
  },
39
  {
40
  "clip_ratio": 1.3441811461234465e-05,
 
44
  "kl": 9.50181856751442e-07,
45
  "learning_rate": 5e-05,
46
  "loss": -0.06604708731174469,
47
+ "memory(GiB)": 182.91,
48
  "response_clip_ratio": 0.13671875,
49
  "reward": 0.0006296975770965219,
50
  "reward_std": 0.07172460854053497,
 
60
  "kl": 1.1101365089416504e-05,
61
  "learning_rate": 6.666666666666667e-05,
62
  "loss": -0.06727766245603561,
63
+ "memory(GiB)": 182.91,
64
  "step": 4,
65
  "train_speed(iter/s)": 0.000458
66
  },
 
72
  "kl": 0.00017762184143066406,
73
  "learning_rate": 8.333333333333334e-05,
74
  "loss": -0.09315311908721924,
75
+ "memory(GiB)": 182.91,
76
  "response_clip_ratio": 0.119140625,
77
  "reward": -0.005135859013535082,
78
  "reward_std": 0.07994875870645046,
 
86
  "grad_norm": 0.18263348937034607,
87
  "learning_rate": 0.0001,
88
  "loss": -0.1041698157787323,
89
+ "memory(GiB)": 182.91,
90
  "step": 6,
91
+ "train_speed(iter/s)": 0.000459
92
  },
93
  {
94
  "epoch": 1.4210526315789473,
 
101
  "eval_reward_std": 0.08769983053207397,
102
  "eval_rewards/CosineReward": 0.012996694073081017,
103
  "eval_rewards/RepetitionPenalty": 0.0,
104
+ "eval_runtime": 1030.1127,
105
  "eval_samples_per_second": 0.001,
106
  "eval_steps_per_second": 0.001,
107
  "step": 6
 
114
  "kl": 0.017406463623046875,
115
  "learning_rate": 9.991540791356342e-05,
116
  "loss": -0.051375165581703186,
117
+ "memory(GiB)": 182.91,
118
  "response_clip_ratio": 0.1484375,
119
  "reward": 0.004909618757665157,
120
  "reward_std": 0.08167182095348835,
 
130
  "kl": 0.089599609375,
131
  "learning_rate": 9.966191788709716e-05,
132
  "loss": -0.05105742812156677,
133
+ "memory(GiB)": 182.91,
134
  "step": 8,
135
  "train_speed(iter/s)": 0.000433
136
  },
 
142
  "kl": 0.0963134765625,
143
  "learning_rate": 9.924038765061042e-05,
144
  "loss": -0.05842069163918495,
145
+ "memory(GiB)": 182.91,
146
  "response_clip_ratio": 0.255859375,
147
  "reward": 0.03643610421568155,
148
  "reward_std": 0.11898956261575222,
 
158
  "kl": 0.1185302734375,
159
  "learning_rate": 9.865224352899119e-05,
160
  "loss": -0.06491819024085999,
161
+ "memory(GiB)": 182.91,
162
  "step": 10,
163
  "train_speed(iter/s)": 0.000436
164
  },
 
170
  "kl": 0.1275634765625,
171
  "learning_rate": 9.789947561577445e-05,
172
  "loss": -0.04600231721997261,
173
+ "memory(GiB)": 182.91,
174
  "response_clip_ratio": 0.361328125,
175
  "reward": 0.023204635945148766,
176
  "reward_std": 0.10593634657561779,
 
184
  "grad_norm": 0.05781339108943939,
185
  "learning_rate": 9.698463103929542e-05,
186
  "loss": -0.05069056898355484,
187
+ "memory(GiB)": 182.91,
188
  "step": 12,
189
  "train_speed(iter/s)": 0.000439
190
  },
 
199
  "eval_reward_std": 0.10685288906097412,
200
  "eval_rewards/CosineReward": 0.03234308212995529,
201
  "eval_rewards/RepetitionPenalty": 0.0,
202
+ "eval_runtime": 1025.9041,
203
  "eval_samples_per_second": 0.001,
204
  "eval_steps_per_second": 0.001,
205
  "step": 12
 
212
  "kl": 0.151123046875,
213
  "learning_rate": 9.591080534401371e-05,
214
  "loss": -0.02191038429737091,
215
+ "memory(GiB)": 182.91,
216
  "response_clip_ratio": 0.419921875,
217
  "reward": 0.035983758978545666,
218
  "reward_std": 0.11553369648754597,
 
228
  "kl": 0.169189453125,
229
  "learning_rate": 9.468163201617062e-05,
230
  "loss": -0.022672578692436218,
231
+ "memory(GiB)": 182.91,
232
  "step": 14,
233
  "train_speed(iter/s)": 0.000427
234
  },
 
240
  "kl": 0.166748046875,
241
  "learning_rate": 9.330127018922194e-05,
242
  "loss": -0.059799157083034515,
243
+ "memory(GiB)": 182.91,
244
  "response_clip_ratio": 0.4765625,
245
  "reward": 0.03584331553429365,
246
  "reward_std": 0.11829411797225475,
 
256
  "kl": 0.16748046875,
257
  "learning_rate": 9.177439057064683e-05,
258
  "loss": -0.06071458384394646,
259
+ "memory(GiB)": 182.91,
260
  "step": 16,
261
  "train_speed(iter/s)": 0.000431
262
  },
 
268
  "kl": 0.1787109375,
269
  "learning_rate": 9.01061596377522e-05,
270
  "loss": -0.04504441097378731,
271
+ "memory(GiB)": 182.91,
272
  "response_clip_ratio": 0.5625,
273
  "reward": 0.027318883687257767,
274
  "reward_std": 0.10441224090754986,
 
282
  "grad_norm": 0.005998397711664438,
283
  "learning_rate": 8.83022221559489e-05,
284
  "loss": -0.045487549155950546,
285
+ "memory(GiB)": 182.91,
286
  "step": 18,
287
  "train_speed(iter/s)": 0.000432
288
  },
 
297
  "eval_reward_std": 0.10691346973180771,
298
  "eval_rewards/CosineReward": 0.03729327768087387,
299
  "eval_rewards/RepetitionPenalty": 0.0,
300
+ "eval_runtime": 1041.231,
301
  "eval_samples_per_second": 0.001,
302
  "eval_steps_per_second": 0.001,
303
  "step": 18
 
310
  "kl": 0.1820068359375,
311
  "learning_rate": 8.636868207865244e-05,
312
  "loss": -0.03466903418302536,
313
+ "memory(GiB)": 182.91,
314
  "response_clip_ratio": 0.466796875,
315
  "reward": 0.04069916973821819,
316
  "reward_std": 0.11991005763411522,
 
326
  "kl": 0.19287109375,
327
  "learning_rate": 8.43120818934367e-05,
328
  "loss": -0.03502114117145538,
329
+ "memory(GiB)": 182.91,
330
  "step": 20,
331
  "train_speed(iter/s)": 0.000424
332
  },
 
338
  "kl": 0.17626953125,
339
  "learning_rate": 8.213938048432697e-05,
340
  "loss": -0.008662773296236992,
341
+ "memory(GiB)": 182.91,
342
  "response_clip_ratio": 0.5625,
343
  "reward": 0.04996980866417289,
344
  "reward_std": 0.13849420100450516,
345
  "rewards/CosineReward": 0.049969930201768875,
346
  "rewards/RepetitionPenalty": -1.1864573679076784e-07,
347
  "step": 21,
348
+ "train_speed(iter/s)": 0.000408
349
  },
350
  {
351
  "clip_ratio": 5.869188044016482e-05,
 
354
  "kl": 0.178955078125,
355
  "learning_rate": 7.985792958513931e-05,
356
  "loss": -0.008743642829358578,
357
+ "memory(GiB)": 182.91,
358
  "step": 22,
359
  "train_speed(iter/s)": 0.000426
360
  },
 
366
  "kl": 0.1796875,
367
  "learning_rate": 7.74754489035403e-05,
368
  "loss": -0.03423420712351799,
369
+ "memory(GiB)": 182.91,
370
  "response_clip_ratio": 0.583984375,
371
  "reward": 0.034468831261619925,
372
  "reward_std": 0.11841745302081108,
 
380
  "grad_norm": 0.014131724834442139,
381
  "learning_rate": 7.500000000000001e-05,
382
  "loss": -0.03426633030176163,
383
+ "memory(GiB)": 182.91,
384
  "step": 24,
385
  "train_speed(iter/s)": 0.000427
386
  },
 
395
  "eval_reward_std": 0.10456253588199615,
396
  "eval_rewards/CosineReward": 0.04339282959699631,
397
  "eval_rewards/RepetitionPenalty": 0.0,
398
+ "eval_runtime": 1045.0632,
399
  "eval_samples_per_second": 0.001,
400
  "eval_steps_per_second": 0.001,
401
  "step": 24
 
408
  "kl": 0.1800537109375,
409
  "learning_rate": 7.243995901002312e-05,
410
  "loss": -0.02097315341234207,
411
+ "memory(GiB)": 182.91,
412
  "response_clip_ratio": 0.6171875,
413
  "reward": 0.03010205877944827,
414
  "reward_std": 0.10742511600255966,
 
424
  "kl": 0.18408203125,
425
  "learning_rate": 6.980398830195785e-05,
426
  "loss": -0.02103913575410843,
427
+ "memory(GiB)": 182.91,
428
  "step": 26,
429
  "train_speed(iter/s)": 0.000421
430
  },
 
436
  "kl": 0.174560546875,
437
  "learning_rate": 6.710100716628344e-05,
438
  "loss": -0.03593946248292923,
439
+ "memory(GiB)": 182.91,
440
  "response_clip_ratio": 0.513671875,
441
  "reward": 0.04752760287374258,
442
  "reward_std": 0.14935147762298584,
 
452
  "kl": 0.182373046875,
453
  "learning_rate": 6.434016163555452e-05,
454
  "loss": -0.03595500811934471,
455
+ "memory(GiB)": 182.91,
456
  "step": 28,
457
  "train_speed(iter/s)": 0.000422
458
  }
checkpoint-28/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:09cdf21dfd9faa218b7fd99e3f3dc0ef681c4e3fd3b905e7348f5467b0198044
3
  size 9809
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1207fcb9d91c7deb13a80104f3ca89016b4cff3ef13ebd136ee6320d5a9888bb
3
  size 9809
checkpoint-30/adapter_config.json CHANGED
@@ -24,10 +24,10 @@
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
27
- "up_proj",
28
- "k_proj",
29
  "gate_proj",
 
30
  "o_proj",
 
31
  "q_proj",
32
  "down_proj"
33
  ],
 
24
  "revision": null,
25
  "target_modules": [
26
  "v_proj",
 
 
27
  "gate_proj",
28
+ "k_proj",
29
  "o_proj",
30
+ "up_proj",
31
  "q_proj",
32
  "down_proj"
33
  ],
checkpoint-30/args.json ADDED
@@ -0,0 +1,416 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "/mnt/nvme5n1p1/distill-14b-rl-70",
3
+ "model_type": "deepseek_r1_distill",
4
+ "model_revision": null,
5
+ "task_type": "causal_lm",
6
+ "torch_dtype": "bfloat16",
7
+ "attn_impl": null,
8
+ "num_labels": null,
9
+ "rope_scaling": null,
10
+ "device_map": null,
11
+ "max_memory": {},
12
+ "local_repo_path": null,
13
+ "template": "deepseek_r1",
14
+ "system": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step. Return final answer within \\\\boxed{}.",
15
+ "max_length": 16384,
16
+ "truncation_strategy": "left",
17
+ "max_pixels": null,
18
+ "tools_prompt": "react_en",
19
+ "norm_bbox": null,
20
+ "response_prefix": null,
21
+ "padding_side": "right",
22
+ "loss_scale": "last_round",
23
+ "sequence_parallel_size": 1,
24
+ "use_chat_template": true,
25
+ "template_backend": "swift",
26
+ "dataset": [
27
+ "stage2_aime.jsonl"
28
+ ],
29
+ "val_dataset": [],
30
+ "split_dataset_ratio": 0.01,
31
+ "data_seed": 42,
32
+ "dataset_num_proc": 52,
33
+ "streaming": false,
34
+ "enable_cache": false,
35
+ "download_mode": "reuse_dataset_if_exists",
36
+ "columns": {},
37
+ "strict": false,
38
+ "remove_unused_columns": false,
39
+ "model_name": [
40
+ null,
41
+ null
42
+ ],
43
+ "model_author": [
44
+ null,
45
+ null
46
+ ],
47
+ "custom_dataset_info": [],
48
+ "quant_method": null,
49
+ "quant_bits": null,
50
+ "hqq_axis": null,
51
+ "bnb_4bit_compute_dtype": "bfloat16",
52
+ "bnb_4bit_quant_type": "nf4",
53
+ "bnb_4bit_use_double_quant": true,
54
+ "bnb_4bit_quant_storage": null,
55
+ "max_new_tokens": 64,
56
+ "temperature": 1.0,
57
+ "top_k": 50,
58
+ "top_p": 0.9,
59
+ "repetition_penalty": 1.1,
60
+ "num_beams": 1,
61
+ "stream": false,
62
+ "stop_words": [],
63
+ "logprobs": false,
64
+ "top_logprobs": null,
65
+ "ckpt_dir": "/mnt/nvme5n1p1/distill-14b-rl-70",
66
+ "load_dataset_config": null,
67
+ "lora_modules": [],
68
+ "tuner_backend": "peft",
69
+ "train_type": "lora",
70
+ "adapters": [],
71
+ "external_plugins": [],
72
+ "seed": 42,
73
+ "model_kwargs": {},
74
+ "load_args": false,
75
+ "load_data_args": false,
76
+ "use_hf": true,
77
+ "hub_token": null,
78
+ "custom_register_path": [],
79
+ "ignore_args_error": false,
80
+ "use_swift_lora": false,
81
+ "output_dir": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345",
82
+ "overwrite_output_dir": false,
83
+ "do_train": false,
84
+ "do_eval": false,
85
+ "do_predict": false,
86
+ "eval_strategy": "steps",
87
+ "prediction_loss_only": false,
88
+ "per_device_train_batch_size": 4,
89
+ "per_device_eval_batch_size": 4,
90
+ "per_gpu_train_batch_size": null,
91
+ "per_gpu_eval_batch_size": null,
92
+ "gradient_accumulation_steps": 4,
93
+ "eval_accumulation_steps": null,
94
+ "eval_delay": 0,
95
+ "torch_empty_cache_steps": null,
96
+ "learning_rate": 0.0001,
97
+ "weight_decay": 0.1,
98
+ "adam_beta1": 0.9,
99
+ "adam_beta2": 0.999,
100
+ "adam_epsilon": 1e-08,
101
+ "max_grad_norm": 1.0,
102
+ "num_train_epochs": 15.0,
103
+ "max_steps": -1,
104
+ "lr_scheduler_type": "cosine",
105
+ "lr_scheduler_kwargs": null,
106
+ "warmup_ratio": 0.1,
107
+ "warmup_steps": 0,
108
+ "log_level": "passive",
109
+ "log_level_replica": "warning",
110
+ "log_on_each_node": true,
111
+ "logging_dir": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/runs",
112
+ "logging_strategy": "steps",
113
+ "logging_first_step": true,
114
+ "logging_steps": 1,
115
+ "logging_nan_inf_filter": true,
116
+ "save_strategy": "steps",
117
+ "save_steps": 2.0,
118
+ "save_total_limit": 100,
119
+ "save_safetensors": true,
120
+ "save_on_each_node": true,
121
+ "save_only_model": false,
122
+ "restore_callback_states_from_checkpoint": false,
123
+ "no_cuda": false,
124
+ "use_cpu": false,
125
+ "use_mps_device": false,
126
+ "jit_mode_eval": false,
127
+ "use_ipex": false,
128
+ "bf16": true,
129
+ "fp16": false,
130
+ "fp16_opt_level": "O1",
131
+ "half_precision_backend": "auto",
132
+ "bf16_full_eval": false,
133
+ "fp16_full_eval": false,
134
+ "tf32": null,
135
+ "local_rank": 0,
136
+ "ddp_backend": null,
137
+ "tpu_num_cores": null,
138
+ "tpu_metrics_debug": false,
139
+ "debug": null,
140
+ "dataloader_drop_last": false,
141
+ "eval_steps": 6.0,
142
+ "dataloader_num_workers": 52,
143
+ "dataloader_prefetch_factor": null,
144
+ "past_index": -1,
145
+ "run_name": null,
146
+ "disable_tqdm": null,
147
+ "label_names": null,
148
+ "load_best_model_at_end": false,
149
+ "metric_for_best_model": "reward",
150
+ "greater_is_better": true,
151
+ "ignore_data_skip": false,
152
+ "fsdp": "",
153
+ "fsdp_min_num_params": 0,
154
+ "fsdp_config": null,
155
+ "fsdp_transformer_layer_cls_to_wrap": null,
156
+ "accelerator_config": {
157
+ "dispatch_batches": false
158
+ },
159
+ "deepspeed": {
160
+ "fp16": {
161
+ "enabled": "auto",
162
+ "loss_scale": 0,
163
+ "loss_scale_window": 1000,
164
+ "initial_scale_power": 16,
165
+ "hysteresis": 2,
166
+ "min_loss_scale": 1
167
+ },
168
+ "bf16": {
169
+ "enabled": "auto"
170
+ },
171
+ "zero_optimization": {
172
+ "stage": 2,
173
+ "offload_optimizer": {
174
+ "device": "none",
175
+ "pin_memory": true
176
+ },
177
+ "allgather_partitions": true,
178
+ "allgather_bucket_size": 200000000.0,
179
+ "overlap_comm": true,
180
+ "reduce_scatter": true,
181
+ "reduce_bucket_size": 200000000.0,
182
+ "contiguous_gradients": true
183
+ },
184
+ "gradient_accumulation_steps": "auto",
185
+ "gradient_clipping": "auto",
186
+ "steps_per_print": 2000,
187
+ "train_batch_size": "auto",
188
+ "train_micro_batch_size_per_gpu": "auto",
189
+ "wall_clock_breakdown": false
190
+ },
191
+ "label_smoothing_factor": 0.0,
192
+ "optim": "adamw_torch",
193
+ "optim_args": null,
194
+ "adafactor": false,
195
+ "group_by_length": false,
196
+ "length_column_name": "length",
197
+ "report_to": [
198
+ "wandb"
199
+ ],
200
+ "ddp_find_unused_parameters": null,
201
+ "ddp_bucket_cap_mb": null,
202
+ "ddp_broadcast_buffers": null,
203
+ "dataloader_pin_memory": true,
204
+ "dataloader_persistent_workers": false,
205
+ "skip_memory_metrics": true,
206
+ "use_legacy_prediction_loop": false,
207
+ "push_to_hub": false,
208
+ "resume_from_checkpoint": null,
209
+ "hub_model_id": null,
210
+ "hub_strategy": "every_save",
211
+ "hub_private_repo": null,
212
+ "hub_always_push": false,
213
+ "gradient_checkpointing": true,
214
+ "gradient_checkpointing_kwargs": null,
215
+ "include_inputs_for_metrics": false,
216
+ "include_for_metrics": [],
217
+ "eval_do_concat_batches": true,
218
+ "fp16_backend": "auto",
219
+ "evaluation_strategy": "steps",
220
+ "push_to_hub_model_id": null,
221
+ "push_to_hub_organization": null,
222
+ "push_to_hub_token": null,
223
+ "mp_parameters": "",
224
+ "auto_find_batch_size": false,
225
+ "full_determinism": false,
226
+ "torchdynamo": null,
227
+ "ray_scope": "last",
228
+ "ddp_timeout": 1800,
229
+ "torch_compile": false,
230
+ "torch_compile_backend": null,
231
+ "torch_compile_mode": null,
232
+ "dispatch_batches": null,
233
+ "split_batches": null,
234
+ "include_tokens_per_second": false,
235
+ "include_num_input_tokens_seen": false,
236
+ "neftune_noise_alpha": null,
237
+ "optim_target_modules": null,
238
+ "batch_eval_metrics": false,
239
+ "eval_on_start": false,
240
+ "use_liger_kernel": false,
241
+ "eval_use_gather_object": false,
242
+ "average_tokens_across_devices": false,
243
+ "sortish_sampler": false,
244
+ "predict_with_generate": false,
245
+ "generation_max_length": null,
246
+ "generation_num_beams": null,
247
+ "generation_config": null,
248
+ "freeze_parameters": [],
249
+ "freeze_parameters_ratio": 0.0,
250
+ "trainable_parameters": [],
251
+ "freeze_llm": false,
252
+ "freeze_vit": true,
253
+ "freeze_aligner": true,
254
+ "target_modules": [
255
+ "all-linear"
256
+ ],
257
+ "target_regex": null,
258
+ "modules_to_save": [],
259
+ "lora_rank": 32,
260
+ "lora_alpha": 32,
261
+ "lora_dropout": 0.05,
262
+ "lora_bias": "none",
263
+ "lora_dtype": null,
264
+ "lorap_lr_ratio": null,
265
+ "use_rslora": false,
266
+ "use_dora": false,
267
+ "lora_ga_batch_size": 2,
268
+ "lora_ga_iters": 2,
269
+ "lora_ga_max_length": 1024,
270
+ "lora_ga_direction": "ArB2r",
271
+ "lora_ga_scale": "stable",
272
+ "lora_ga_stable_gamma": 16,
273
+ "init_weights": true,
274
+ "fourier_n_frequency": 2000,
275
+ "fourier_scaling": 300.0,
276
+ "boft_block_size": 4,
277
+ "boft_block_num": 0,
278
+ "boft_n_butterfly_factor": 1,
279
+ "boft_dropout": 0.0,
280
+ "vera_rank": 256,
281
+ "vera_projection_prng_key": 0,
282
+ "vera_dropout": 0.0,
283
+ "vera_d_initial": 0.1,
284
+ "adapter_act": "gelu",
285
+ "adapter_length": 128,
286
+ "use_galore": false,
287
+ "galore_target_modules": null,
288
+ "galore_rank": 128,
289
+ "galore_update_proj_gap": 50,
290
+ "galore_scale": 1.0,
291
+ "galore_proj_type": "std",
292
+ "galore_optim_per_parameter": false,
293
+ "galore_with_embedding": false,
294
+ "galore_quantization": false,
295
+ "galore_proj_quant": false,
296
+ "galore_proj_bits": 4,
297
+ "galore_proj_group_size": 256,
298
+ "galore_cos_threshold": 0.4,
299
+ "galore_gamma_proj": 2,
300
+ "galore_queue_size": 5,
301
+ "adalora_target_r": 8,
302
+ "adalora_init_r": 12,
303
+ "adalora_tinit": 0,
304
+ "adalora_tfinal": 0,
305
+ "adalora_deltaT": 1,
306
+ "adalora_beta1": 0.85,
307
+ "adalora_beta2": 0.85,
308
+ "adalora_orth_reg_weight": 0.5,
309
+ "llamapro_num_new_blocks": 4,
310
+ "llamapro_num_groups": null,
311
+ "lisa_activated_layers": 0,
312
+ "lisa_step_interval": 20,
313
+ "reft_layer_key": null,
314
+ "reft_layers": null,
315
+ "reft_rank": 4,
316
+ "reft_intervention_type": "LoreftIntervention",
317
+ "reft_args": null,
318
+ "use_liger": false,
319
+ "model_layer_cls_name": null,
320
+ "metric_warmup_step": 0,
321
+ "fsdp_num": 1,
322
+ "acc_steps": 1,
323
+ "swanlab_token": null,
324
+ "swanlab_project": null,
325
+ "swanlab_workspace": null,
326
+ "swanlab_exp_name": null,
327
+ "swanlab_mode": "cloud",
328
+ "add_version": true,
329
+ "resume_only_model": false,
330
+ "check_model": true,
331
+ "create_checkpoint_symlink": false,
332
+ "packing": false,
333
+ "lazy_tokenize": false,
334
+ "loss_type": null,
335
+ "optimizer": null,
336
+ "metric": null,
337
+ "acc_strategy": "token",
338
+ "zero_hpz_partition_size": null,
339
+ "reward_model": null,
340
+ "reward_adapters": [],
341
+ "reward_model_type": null,
342
+ "reward_model_revision": null,
343
+ "num_ppo_epochs": 4,
344
+ "whiten_rewards": false,
345
+ "kl_coef": 0.05,
346
+ "cliprange": 0.2,
347
+ "vf_coef": 0.1,
348
+ "cliprange_value": 0.2,
349
+ "gamma": 1.0,
350
+ "lam": 0.95,
351
+ "num_mini_batches": 1,
352
+ "local_rollout_forward_batch_size": 64,
353
+ "num_sample_generations": 10,
354
+ "response_length": 512,
355
+ "missing_eos_penalty": null,
356
+ "num_infer_workers": 8,
357
+ "vllm_max_num_seqs": 256,
358
+ "vllm_enforce_eager": false,
359
+ "vllm_limit_mm_per_prompt": null,
360
+ "vllm_enable_prefix_caching": true,
361
+ "cosine_min_len_value_wrong": 0.0,
362
+ "cosine_max_len_value_wrong": -0.2,
363
+ "cosine_min_len_value_correct": 0.8,
364
+ "cosine_max_len_value_correct": 0.4,
365
+ "cosine_max_len": 12288,
366
+ "repetition_n_grams": 40,
367
+ "repetition_max_penalty": -0.05,
368
+ "use_lmdeploy": false,
369
+ "lmdeploy_device": "auto",
370
+ "lmdeploy_session_len": null,
371
+ "lmdeploy_cache_max_entry_count": 0.8,
372
+ "async_generate": false,
373
+ "tensor_parallel_size": 1,
374
+ "sleep_level": 0,
375
+ "move_model_batches": null,
376
+ "offload_optimizer": false,
377
+ "offload_model": false,
378
+ "gc_collect_after_offload": false,
379
+ "num_generations": 8,
380
+ "max_completion_length": 12288,
381
+ "ds3_gather_for_generation": true,
382
+ "reward_funcs": [
383
+ "cosine",
384
+ "repetition"
385
+ ],
386
+ "reward_weights": null,
387
+ "log_completions": true,
388
+ "use_vllm": false,
389
+ "vllm_device": [
390
+ "auto"
391
+ ],
392
+ "vllm_gpu_memory_utilization": 0.9,
393
+ "vllm_max_model_len": null,
394
+ "num_iterations": 2,
395
+ "epsilon": 0.2,
396
+ "rlhf_type": "grpo",
397
+ "ref_model": null,
398
+ "ref_model_type": null,
399
+ "ref_model_revision": null,
400
+ "beta": 0.04,
401
+ "label_smoothing": 0,
402
+ "rpo_alpha": 1.0,
403
+ "cpo_alpha": 1.0,
404
+ "simpo_gamma": 1,
405
+ "desirable_weight": 1.0,
406
+ "undesirable_weight": 1.0,
407
+ "rank": 0,
408
+ "global_world_size": 32,
409
+ "local_world_size": 8,
410
+ "model_suffix": "distill-14b-rl-70",
411
+ "model_info": "ModelInfo(model_type='deepseek_r1_distill', model_dir='/mnt/nvme5n1p1/distill-14b-rl-70', torch_dtype=torch.bfloat16, max_model_len=131072, quant_method=None, quant_bits=None, rope_scaling=None, config=None, task_type='causal_lm', num_labels=None)",
412
+ "model_meta": "ModelMeta(model_type='deepseek_r1_distill', model_groups=[ModelGroup(models=[Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-7B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-7B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-14B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-14B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-32B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-32B', model_path=None, ms_revision=None, hf_revision=None)], ignore_patterns=None, requires=['transformers>=4.37'], tags=[]), ModelGroup(models=[Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Llama-8B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Llama-8B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='deepseek-ai/DeepSeek-R1-Distill-Llama-70B', hf_model_id='deepseek-ai/DeepSeek-R1-Distill-Llama-70B', model_path=None, ms_revision=None, hf_revision=None)], ignore_patterns=None, requires=None, tags=[])], template='deepseek_r1', get_function=<function get_model_tokenizer_with_flash_attn at 0x7f57e39d3400>, model_arch='llama', architectures=['Qwen2ForCausalLM', 'LlamaForCausalLM'], additional_saved_files=[], torch_dtype=None, is_multimodal=False, is_reward=False, task_type=None, ignore_patterns=[], requires=[], tags=[])",
413
+ "model_dir": "/mnt/nvme5n1p1/distill-14b-rl-70",
414
+ "hub": "<class 'swift.hub.hub.HFHub'>",
415
+ "training_args": "GRPOConfig(output_dir='/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345', overwrite_output_dir=False, do_train=False, do_eval=True, do_predict=False, eval_strategy=<IntervalStrategy.STEPS: 'steps'>, prediction_loss_only=False, per_device_train_batch_size=4, per_device_eval_batch_size=4, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=4, eval_accumulation_steps=None, eval_delay=0, torch_empty_cache_steps=None, learning_rate=0.0001, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.999, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=15.0, max_steps=-1, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_ratio=0.1, warmup_steps=0, log_level='passive', log_level_replica='warning', log_on_each_node=True, logging_dir='/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/runs', logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_first_step=True, logging_steps=1, logging_nan_inf_filter=True, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=2, save_total_limit=100, save_safetensors=True, save_on_each_node=True, save_only_model=False, restore_callback_states_from_checkpoint=False, no_cuda=False, use_cpu=False, use_mps_device=False, seed=42, data_seed=42, jit_mode_eval=False, use_ipex=False, bf16=True, fp16=False, fp16_opt_level='O1', half_precision_backend='auto', bf16_full_eval=False, fp16_full_eval=False, tf32=None, local_rank=0, ddp_backend=None, tpu_num_cores=None, tpu_metrics_debug=False, debug=[], dataloader_drop_last=False, eval_steps=6, dataloader_num_workers=52, dataloader_prefetch_factor=None, past_index=-1, run_name='/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345', disable_tqdm=False, remove_unused_columns=False, label_names=None, load_best_model_at_end=False, metric_for_best_model='reward', greater_is_better=True, ignore_data_skip=False, fsdp=[], fsdp_min_num_params=0, fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, fsdp_transformer_layer_cls_to_wrap=None, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': True, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, label_smoothing_factor=0.0, optim=<OptimizerNames.ADAMW_TORCH: 'adamw_torch'>, optim_args=None, adafactor=False, group_by_length=False, length_column_name='length', report_to=['wandb'], ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, dataloader_pin_memory=True, dataloader_persistent_workers=False, skip_memory_metrics=True, use_legacy_prediction_loop=False, push_to_hub=False, resume_from_checkpoint=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_token=None, hub_private_repo=None, hub_always_push=False, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, include_inputs_for_metrics=False, include_for_metrics=[], eval_do_concat_batches=True, fp16_backend='auto', evaluation_strategy='steps', push_to_hub_model_id=None, push_to_hub_organization=None, push_to_hub_token=None, mp_parameters='', auto_find_batch_size=False, full_determinism=False, torchdynamo=None, ray_scope='last', ddp_timeout=1800, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, dispatch_batches=None, split_batches=None, include_tokens_per_second=None, include_num_input_tokens_seen=None, neftune_noise_alpha=None, optim_target_modules=None, batch_eval_metrics=False, eval_on_start=False, use_liger_kernel=False, eval_use_gather_object=False, average_tokens_across_devices=None, model_init_kwargs=None, max_prompt_length=512, num_generations=8, max_completion_length=12288, ds3_gather_for_generation=True, temperature=1.0, top_p=0.9, top_k=50, min_p=None, repetition_penalty=1.1, cache_implementation=None, use_vllm=False, vllm_server_host='0.0.0.0', vllm_server_port=8000, vllm_server_timeout=120.0, vllm_guided_decoding_regex=None, beta=0.04, num_iterations=2, epsilon=0.2, epsilon_high=None, reward_weights=None, scale_rewards=True, sync_ref_model=False, ref_model_mixup_alpha=0.6, ref_model_sync_steps=512, log_completions=True, vllm_device=['auto'], vllm_gpu_memory_utilization=0.9, vllm_dtype=None, vllm_max_model_len=None, vllm_enable_prefix_caching=True, acc_strategy='token', sequence_parallel_size=1, check_model=True, train_sampler_random=True, is_encoder_decoder=False, metric_warmup_step=0, train_dataset_sample=-1, fsdp_num=1, acc_steps=1, train_type='lora', optimizer=None, local_repo_path=None, galore_config=None, num_infer_workers=8, vllm_max_num_seqs=256, vllm_enforce_eager=False, vllm_limit_mm_per_prompt={}, cosine_min_len_value_wrong=0.0, cosine_max_len_value_wrong=-0.2, cosine_min_len_value_correct=0.8, cosine_max_len_value_correct=0.4, cosine_max_len=12288, repetition_n_grams=40, repetition_max_penalty=-0.05, use_lmdeploy=False, lmdeploy_device='auto', lmdeploy_session_len=None, lmdeploy_cache_max_entry_count=0.8, async_generate=False, tensor_parallel_size=1, sleep_level=0, move_model_batches=None, offload_optimizer=False, offload_model=False, gc_collect_after_offload=False, stop_words=[])"
416
+ }
checkpoint-30/global_step30/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:116fd5b5622dc5ab41c2b4b69599992b88a191c17075a38cd7dadf69575e4a78
3
+ size 51616517
checkpoint-30/global_step30/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10e8a2acf8c0ac1a9ffa239a43c6bb598040c0220e25be565e2bb53eaa1bd714
3
+ size 51616005
checkpoint-30/global_step30/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30af77056d615b52d15c56ee9753e492629de7963f1d6ee5ef8a338886879edb
3
+ size 51616517
checkpoint-30/global_step30/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ec075b7a811dd204fd12c6ac145a9dceccdc4f56ffc59571a349c2da1de140f
3
+ size 51616005
checkpoint-30/global_step30/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eaec79a6883ccbbfee57c719f45ee94fc319c6a85cf44e731a557599e99e08af
3
+ size 51616517
checkpoint-30/global_step30/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70645416a07a17e9a5d87ba5a084531ba3a59ee64549321dce7acc6591bd3cbd
3
+ size 51616005
checkpoint-30/global_step30/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfe1c45f684c024d69a085a30832b079cedf161331d4d1d2b0ca34e083764488
3
+ size 51616517
checkpoint-30/global_step30/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:81e017409e9f8dd518dadb52434224c2a69a811e9ab613d40be8809afe1b1272
3
+ size 51616005
checkpoint-30/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7445d8a3801c08606f052af6eba1f066506146bc8850e86b93b99b9724d6a06b
3
+ size 16389
checkpoint-30/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:babfc1ea5e5f1bc14b7cccbd566cff5d7cb473de1e0d77762bca88dc3bc2dea3
3
+ size 16389
checkpoint-30/rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a431c5a1142e39a8f6ceca0202b05dafd58fc61cfbf3f69fe6950fa8d0b864ea
3
+ size 16389
checkpoint-30/rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c134753e443e6f990e92a23324f4b3e5487167a3cc2777f478d832bd3af0b587
3
+ size 16389
checkpoint-30/rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e26df1338c58354b929a3988f94ed860e003fcf9b7a805288ea975992aeb1387
3
+ size 16389
checkpoint-30/rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6208c597b8eaeb85184446a7e8fe032bef0bbfdd8a59aa7c20b1bc8e8b007d34
3
+ size 16325
checkpoint-30/rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8803b97be66ed6f3d4987aa0c579ab126c1b7e2347e0dcb094f20dcdccebce5
3
+ size 16389