End of training

README.md

@@ -39,7 +39,7 @@ This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing
 ### Framework versions
 
 - TRL: 0.15.2
-- Transformers: 4.51.3
+- Transformers: 4.48.2
 - Pytorch: 2.5.1
 - Datasets: 3.6.0
 - Tokenizers: 0.21.1

adapter_config.json

@@ -24,13 +24,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
     "up_proj",
+    "down_proj",
     "q_proj",
+    "gate_proj",
     "o_proj",
-    "down_proj",
-    "k_proj"
+    "k_proj",
+    "v_proj"
   ],
   "task_type": "CAUSAL_LM",
   "trainable_token_indices": null,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e72cce6b63b5a77942d08f0640efa6e368a95045eaf7753a8839f9b21346b3d5
+oid sha256:c63792df00daeacb83fd5a80135dadb878374da13204e70e16592264c99a6802
 size 35237104

all_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 2.9286140375006653e-06,
-    "train_runtime": 964.9149,
-    "train_samples": 14,
-    "train_samples_per_second": 0.332,
-    "train_steps_per_second": 0.021
+    "train_loss": 2.333011286737019e-06,
+    "train_runtime": 1452.1013,
+    "train_samples": 19,
+    "train_samples_per_second": 0.22,
+    "train_steps_per_second": 0.014
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 2.9286140375006653e-06,
-    "train_runtime": 964.9149,
-    "train_samples": 14,
-    "train_samples_per_second": 0.332,
-    "train_steps_per_second": 0.021
+    "train_loss": 2.333011286737019e-06,
+    "train_runtime": 1452.1013,
+    "train_samples": 19,
+    "train_samples_per_second": 0.22,
+    "train_steps_per_second": 0.014
 }

trainer_state.json

@@ -1,8 +1,7 @@
 {
-  "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 5.0,
+  "epoch": 4.0,
   "eval_steps": 500,
   "global_step": 20,
   "is_hyper_param_search": false,
@@ -10,209 +9,209 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "completion_length": 425.5625,
-      "epoch": 0.5714285714285714,
-      "grad_norm": 24.163461685180664,
-      "kl": 0.004644934721000027,
-      "learning_rate": 5e-07,
+      "completion_length": 345.78125,
+      "epoch": 0.42105263157894735,
+      "grad_norm": 1.91469407081604,
+      "kl": 0.002231337522971444,
+      "learning_rate": 4.965903258506806e-07,
       "loss": 0.0,
-      "reward": 0.13328281906433403,
-      "reward_std": 0.47804759233258665,
-      "rewards/concensus_correctness_reward_func": 0.0,
+      "reward": 0.29195431526750326,
+      "reward_std": 0.752438448369503,
+      "rewards/concensus_correctness_reward_func": 0.039124999195337296,
       "rewards/consensus_reward_func": 0.0,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.11918907589279115,
+      "rewards/final_correctness_reward_func": 0.0625,
+      "rewards/question_recreation_reward_func": 0.34751682076603174,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.01409374619834125,
+      "rewards/xmlcount_reward_func": -0.15718749910593033,
       "step": 2
     },
     {
-      "completion_length": 382.0833333333333,
-      "epoch": 1.0,
-      "grad_norm": 0.9436554312705994,
-      "kl": 0.002600587273870284,
-      "learning_rate": 4.864543104251586e-07,
+      "completion_length": 349.125,
+      "epoch": 0.8421052631578947,
+      "grad_norm": 1.6588529348373413,
+      "kl": 0.002210008809925057,
+      "learning_rate": 4.698684378016222e-07,
       "loss": 0.0,
-      "reward": 0.17591995932161808,
-      "reward_std": 0.476628802716732,
-      "rewards/concensus_correctness_reward_func": 0.010416666666666666,
+      "reward": 0.7831258047372103,
+      "reward_std": 1.008952866308391,
+      "rewards/concensus_correctness_reward_func": 0.12256249785423279,
       "rewards/consensus_reward_func": 0.0,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.21404495586951575,
+      "rewards/final_correctness_reward_func": 0.375,
+      "rewards/question_recreation_reward_func": 0.2274695634841919,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": -0.048541667560736336,
+      "rewards/xmlcount_reward_func": 0.0580937466584146,
       "step": 4
     },
     {
-      "completion_length": 333.34375,
-      "epoch": 1.5714285714285714,
-      "grad_norm": 1.889753818511963,
-      "kl": 0.0014919452514732257,
-      "learning_rate": 4.472851273490984e-07,
+      "completion_length": 402.57142857142856,
+      "epoch": 1.2105263157894737,
+      "grad_norm": 2.6808555126190186,
+      "kl": 0.003944302243845803,
+      "learning_rate": 4.193203929064353e-07,
       "loss": 0.0,
-      "reward": 0.17828756012022495,
-      "reward_std": 0.22807068983092904,
-      "rewards/concensus_correctness_reward_func": 0.0,
-      "rewards/consensus_reward_func": 0.0,
+      "reward": 1.028327831200191,
+      "reward_std": 1.1498553667749678,
+      "rewards/concensus_correctness_reward_func": 0.14014285802841187,
+      "rewards/consensus_reward_func": 0.07142857142857142,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.14210005290806293,
+      "rewards/final_correctness_reward_func": 0.21428571428571427,
+      "rewards/question_recreation_reward_func": 0.3410063875573022,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.036187499994412065,
+      "rewards/xmlcount_reward_func": 0.2614642892565046,
       "step": 6
     },
     {
-      "completion_length": 406.5,
-      "epoch": 2.0,
-      "grad_norm": 45.66042709350586,
-      "kl": 0.003328462829813361,
-      "learning_rate": 3.867370395306068e-07,
+      "completion_length": 319.4375,
+      "epoch": 1.631578947368421,
+      "grad_norm": 4.16013765335083,
+      "kl": 0.0022768215858377516,
+      "learning_rate": 3.5042385616324236e-07,
       "loss": 0.0,
-      "reward": 0.4213152602314949,
-      "reward_std": 0.4277126689751943,
-      "rewards/concensus_correctness_reward_func": 0.010500000168879827,
-      "rewards/consensus_reward_func": 0.0,
+      "reward": 1.5005913376808167,
+      "reward_std": 2.108041614294052,
+      "rewards/concensus_correctness_reward_func": 0.7163750007748604,
+      "rewards/consensus_reward_func": 0.0625,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.2647319386402766,
+      "rewards/final_correctness_reward_func": 0.125,
+      "rewards/question_recreation_reward_func": 0.3670601084595546,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.14608333073556423,
+      "rewards/xmlcount_reward_func": 0.22965625021606684,
       "step": 8
     },
     {
-      "completion_length": 385.0,
-      "epoch": 2.571428571428571,
-      "grad_norm": 2.002534866333008,
-      "kl": 0.0030680414129165,
-      "learning_rate": 3.1137137178519977e-07,
+      "completion_length": 285.0,
+      "epoch": 2.0,
+      "grad_norm": 2.294445514678955,
+      "kl": 0.0026452127578003065,
+      "learning_rate": 2.706448363680831e-07,
       "loss": 0.0,
-      "reward": 0.20677403640002012,
-      "reward_std": 0.6416971213184297,
-      "rewards/concensus_correctness_reward_func": 0.0078125,
+      "reward": 1.4123743091310774,
+      "reward_std": 1.01517368214471,
+      "rewards/concensus_correctness_reward_func": 0.13742856894220626,
       "rewards/consensus_reward_func": 0.0,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.2949927975423634,
+      "rewards/final_correctness_reward_func": 0.7142857142857143,
+      "rewards/question_recreation_reward_func": 0.3533743140952928,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.015625,
-      "rewards/xmlcount_reward_func": -0.11165625578723848,
+      "rewards/strict_format_reward_func": 0.0,
+      "rewards/xmlcount_reward_func": 0.2072857107434954,
       "step": 10
     },
     {
-      "completion_length": 325.0833333333333,
-      "epoch": 3.0,
-      "grad_norm": 1.308306336402893,
-      "kl": 0.008329887292347848,
-      "learning_rate": 2.2935516363191693e-07,
+      "completion_length": 369.0,
+      "epoch": 2.4210526315789473,
+      "grad_norm": 3.066178560256958,
+      "kl": 0.002823778530000709,
+      "learning_rate": 1.886286282148002e-07,
       "loss": 0.0,
-      "reward": 0.31720937648788095,
-      "reward_std": 0.37778538403411704,
+      "reward": 0.9884280562400818,
+      "reward_std": 1.076359886676073,
       "rewards/concensus_correctness_reward_func": 0.0,
       "rewards/consensus_reward_func": 0.0,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.3178760432638228,
+      "rewards/final_correctness_reward_func": 0.4375,
+      "rewards/question_recreation_reward_func": 0.3941780310124159,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": -0.0006666667759418488,
+      "rewards/xmlcount_reward_func": 0.15675000147894025,
       "step": 12
     },
     {
-      "completion_length": 491.71875,
-      "epoch": 3.571428571428571,
-      "grad_norm": 1.7765281200408936,
-      "kl": 0.0017257212602999061,
-      "learning_rate": 1.4957614383675767e-07,
+      "completion_length": 258.8125,
+      "epoch": 2.8421052631578947,
+      "grad_norm": 4.03523063659668,
+      "kl": 0.0028213907207828015,
+      "learning_rate": 1.1326296046939333e-07,
       "loss": 0.0,
-      "reward": 0.1340795997530222,
-      "reward_std": 0.2006714369636029,
-      "rewards/concensus_correctness_reward_func": 0.0,
-      "rewards/consensus_reward_func": 0.0,
+      "reward": 1.5553401838988066,
+      "reward_std": 2.0138666601851583,
+      "rewards/concensus_correctness_reward_func": 0.7452500015497208,
+      "rewards/consensus_reward_func": 0.0625,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.11620460543781519,
+      "rewards/final_correctness_reward_func": 0.1875,
+      "rewards/question_recreation_reward_func": 0.30774645414203405,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.017874995712190866,
+      "rewards/xmlcount_reward_func": 0.25234374590218067,
       "step": 14
     },
     {
-      "completion_length": 320.5833333333333,
-      "epoch": 4.0,
-      "grad_norm": 0.9761248826980591,
-      "kl": 0.005105598461038123,
-      "learning_rate": 8.067960709356478e-08,
+      "completion_length": 310.35714285714283,
+      "epoch": 3.2105263157894735,
+      "grad_norm": 3.178826093673706,
+      "kl": 0.001912726937527103,
+      "learning_rate": 5.271487265090163e-08,
       "loss": 0.0,
-      "reward": 0.0632177044947942,
-      "reward_std": 0.43164925277233124,
-      "rewards/concensus_correctness_reward_func": 0.0,
-      "rewards/consensus_reward_func": 0.0,
+      "reward": 0.4004705489746162,
+      "reward_std": 1.051499856369836,
+      "rewards/concensus_correctness_reward_func": 0.06871428659984044,
+      "rewards/consensus_reward_func": 0.07142857142857142,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.09588436728032927,
+      "rewards/final_correctness_reward_func": 0.07142857142857142,
+      "rewards/question_recreation_reward_func": 0.21786337984459742,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": -0.032666667054096855,
+      "rewards/strict_format_reward_func": 0.017857142857142856,
+      "rewards/xmlcount_reward_func": -0.04682143724390438,
       "step": 16
     },
     {
-      "completion_length": 454.84375,
-      "epoch": 4.571428571428571,
-      "grad_norm": 2.150505304336548,
-      "kl": 0.0015304867119994015,
-      "learning_rate": 3.013156219837776e-08,
+      "completion_length": 413.9375,
+      "epoch": 3.6315789473684212,
+      "grad_norm": 6.53667688369751,
+      "kl": 0.0016312935331370682,
+      "learning_rate": 1.3545689574841341e-08,
       "loss": 0.0,
-      "reward": 0.3503156821243465,
-      "reward_std": 0.5320424800738692,
+      "reward": 0.5633416045457125,
+      "reward_std": 0.9396487567573786,
       "rewards/concensus_correctness_reward_func": 0.0,
       "rewards/consensus_reward_func": 0.0,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.28294066758826375,
+      "rewards/final_correctness_reward_func": 0.1875,
+      "rewards/question_recreation_reward_func": 0.36087287683039904,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.067375005222857,
+      "rewards/xmlcount_reward_func": 0.014968743547797203,
       "step": 18
     },
     {
-      "completion_length": 396.4583333333333,
-      "epoch": 5.0,
-      "grad_norm": 31.464075088500977,
-      "kl": 0.0031532842840533704,
-      "learning_rate": 3.4096741493194193e-09,
+      "completion_length": 340.4642857142857,
+      "epoch": 4.0,
+      "grad_norm": 1.7881169319152832,
+      "kl": 0.0021190502946930273,
+      "learning_rate": 0.0,
       "loss": 0.0,
-      "reward": 0.18347880616784096,
-      "reward_std": 0.3113321668157975,
-      "rewards/concensus_correctness_reward_func": 0.0,
-      "rewards/consensus_reward_func": 0.0,
+      "reward": 0.9210910541670663,
+      "reward_std": 1.4995469110352653,
+      "rewards/concensus_correctness_reward_func": 0.1758571437426976,
+      "rewards/consensus_reward_func": 0.07142857142857142,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.0,
-      "rewards/question_recreation_reward_func": 0.151103804508845,
+      "rewards/final_correctness_reward_func": 0.2857142857142857,
+      "rewards/question_recreation_reward_func": 0.30434104667178224,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.03237500041723251,
+      "rewards/xmlcount_reward_func": 0.083749991442476,
       "step": 20
     },
     {
-      "epoch": 5.0,
+      "epoch": 4.0,
       "step": 20,
       "total_flos": 0.0,
-      "train_loss": 2.9286140375006653e-06,
-      "train_runtime": 964.9149,
-      "train_samples_per_second": 0.332,
-      "train_steps_per_second": 0.021
+      "train_loss": 2.333011286737019e-06,
+      "train_runtime": 1452.1013,
+      "train_samples_per_second": 0.22,
+      "train_steps_per_second": 0.014
     }
   ],
   "logging_steps": 2,
   "max_steps": 20,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 7,
+  "num_train_epochs": 5,
   "save_steps": 25,
   "stateful_callbacks": {
     "TrainerControl": {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:937cd7c28d3bce6f5912bca51d8187ca6d26db5a3647d13c80d83e727215e159
-size 6008
+oid sha256:200f79e4118f3cb633aa4ee244d119a952e98716de9dfaa52652ffc9a9bc2135
+size 5944