Training in progress, step 800

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b476e4eabd4363d09b656cbae6b34ef5fa2e7a8b0c1589d6700cb9974071d09
+oid sha256:6ee410d6ad8c8069856d52ec95a5882b019aad34c9c6ad64da0ca7c8bbaa05c5
 size 319977229

checkpoint-600/adapter_model/adapter_model/README.md

@@ -4,6 +4,17 @@ library_name: peft
 ## Training procedure
 
 
+The following `bitsandbytes` quantization config was used during training:
+- load_in_8bit: False
+- load_in_4bit: True
+- llm_int8_threshold: 6.0
+- llm_int8_skip_modules: None
+- llm_int8_enable_fp32_cpu_offload: False
+- llm_int8_has_fp16_weight: False
+- bnb_4bit_quant_type: nf4
+- bnb_4bit_use_double_quant: True
+- bnb_4bit_compute_dtype: bfloat16
+
 The following `bitsandbytes` quantization config was used during training:
 - load_in_8bit: False
 - load_in_4bit: True
@@ -16,5 +27,6 @@ The following `bitsandbytes` quantization config was used during training:
 - bnb_4bit_compute_dtype: bfloat16
 ### Framework versions
 
+- PEFT 0.4.0
 
 - PEFT 0.4.0

checkpoint-600/adapter_model/adapter_model/adapter_config.json

@@ -14,13 +14,13 @@
   "r": 64,
   "revision": null,
   "target_modules": [
-    "v_proj",
-    "down_proj",
-    "gate_proj",
     "up_proj",
+    "q_proj",
     "k_proj",
-    "o_proj",
-    "q_proj"
+    "down_proj",
+    "gate_proj",
+    "v_proj",
+    "o_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

checkpoint-600/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51bb9bdb2be0f9d294eae865558a7ead86de7b1ef118aa1dd2a4af1d6edb5d9a
+oid sha256:0b476e4eabd4363d09b656cbae6b34ef5fa2e7a8b0c1589d6700cb9974071d09
 size 319977229

checkpoint-800/README.md

@@ -4,6 +4,17 @@ library_name: peft
 ## Training procedure
 
 
+The following `bitsandbytes` quantization config was used during training:
+- load_in_8bit: False
+- load_in_4bit: True
+- llm_int8_threshold: 6.0
+- llm_int8_skip_modules: None
+- llm_int8_enable_fp32_cpu_offload: False
+- llm_int8_has_fp16_weight: False
+- bnb_4bit_quant_type: nf4
+- bnb_4bit_use_double_quant: True
+- bnb_4bit_compute_dtype: bfloat16
+
 The following `bitsandbytes` quantization config was used during training:
 - load_in_8bit: False
 - load_in_4bit: True
@@ -16,5 +27,6 @@ The following `bitsandbytes` quantization config was used during training:
 - bnb_4bit_compute_dtype: bfloat16
 ### Framework versions
 
+- PEFT 0.4.0
 
 - PEFT 0.4.0

checkpoint-800/adapter_config.json

@@ -14,13 +14,13 @@
   "r": 64,
   "revision": null,
   "target_modules": [
-    "v_proj",
-    "down_proj",
-    "gate_proj",
     "up_proj",
+    "q_proj",
     "k_proj",
-    "o_proj",
-    "q_proj"
+    "down_proj",
+    "gate_proj",
+    "v_proj",
+    "o_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

checkpoint-800/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eab1e2ae5e089eaff970cc67937b7c2fbcd12801c9fc2a5e55e3dc91c4282898
+oid sha256:6ee410d6ad8c8069856d52ec95a5882b019aad34c9c6ad64da0ca7c8bbaa05c5
 size 319977229

checkpoint-800/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14a4eccf4ce423e940f831700ccf138e04b53db97e6d414e28bd3fe42da2f968
+oid sha256:18a1f66ab64d0c0beca4d4f28cb834af27b3c1f1854be1aaa4b006d11064ded6
 size 1279539973

checkpoint-800/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b40f67bf3e80ab79626955f0cf7d8c527bc18e34ff0bc9beda86294caa511eb4
+oid sha256:793ba559bc7ae22f1efbd5c0502afa70d1cadccc9e28a0cc94e4d1cca6ebe596
 size 14511

checkpoint-800/trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.8209574222564697,
+  "best_metric": 0.8205058574676514,
   "best_model_checkpoint": "experts/expert-22/checkpoint-800",
   "epoch": 0.3079588105090944,
   "global_step": 800,
@@ -10,13 +10,13 @@
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.9402,
+      "loss": 0.9409,
       "step": 10
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.9159,
+      "loss": 0.9155,
       "step": 20
     },
     {
@@ -28,25 +28,25 @@
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8795,
+      "loss": 0.8794,
       "step": 40
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.9281,
+      "loss": 0.9277,
       "step": 50
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8639,
+      "loss": 0.864,
       "step": 60
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.9537,
+      "loss": 0.953,
       "step": 70
     },
     {
@@ -64,246 +64,246 @@
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8651,
+      "loss": 0.8647,
       "step": 100
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.9332,
+      "loss": 0.9335,
       "step": 110
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8883,
+      "loss": 0.8879,
       "step": 120
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8945,
+      "loss": 0.8953,
       "step": 130
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8675,
+      "loss": 0.868,
       "step": 140
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8583,
+      "loss": 0.8585,
       "step": 150
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8992,
+      "loss": 0.8998,
       "step": 160
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8667,
+      "loss": 0.8672,
       "step": 170
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8367,
+      "loss": 0.8373,
       "step": 180
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8495,
+      "loss": 0.8492,
       "step": 190
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8338,
+      "loss": 0.8342,
       "step": 200
     },
     {
       "epoch": 0.08,
-      "eval_loss": 0.8420786261558533,
-      "eval_runtime": 104.6029,
-      "eval_samples_per_second": 9.56,
-      "eval_steps_per_second": 4.78,
+      "eval_loss": 0.8420215845108032,
+      "eval_runtime": 77.6159,
+      "eval_samples_per_second": 12.884,
+      "eval_steps_per_second": 6.442,
       "step": 200
     },
     {
       "epoch": 0.08,
-      "mmlu_eval_accuracy": 0.49890909250031423,
+      "mmlu_eval_accuracy": 0.5020690882678942,
       "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
       "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.5862068965517241,
-      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.5,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
       "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
       "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
-      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
       "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
-      "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
       "mmlu_eval_accuracy_global_facts": 0.6,
-      "mmlu_eval_accuracy_high_school_biology": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.53125,
       "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
       "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
-      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
       "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
       "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
       "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
       "mmlu_eval_accuracy_human_aging": 0.782608695652174,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
-      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
       "mmlu_eval_accuracy_machine_learning": 0.45454545454545453,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
-      "mmlu_eval_accuracy_marketing": 0.88,
+      "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
       "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
       "mmlu_eval_accuracy_moral_scenarios": 0.21,
-      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
-      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
       "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
-      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
-      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
       "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
-      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
       "mmlu_eval_accuracy_sociology": 0.7272727272727273,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.4345150615621174,
+      "mmlu_loss": 1.42281840854463,
       "step": 200
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8627,
+      "loss": 0.8625,
       "step": 210
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8758,
+      "loss": 0.8763,
       "step": 220
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.929,
+      "loss": 0.9287,
       "step": 230
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.8946,
+      "loss": 0.8947,
       "step": 240
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.9027,
+      "loss": 0.9023,
       "step": 250
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.866,
+      "loss": 0.8663,
       "step": 260
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.8648,
+      "loss": 0.8644,
       "step": 270
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8545,
+      "loss": 0.8549,
       "step": 280
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8552,
+      "loss": 0.8546,
       "step": 290
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8473,
+      "loss": 0.8476,
       "step": 300
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8627,
+      "loss": 0.8623,
       "step": 310
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8369,
+      "loss": 0.836,
       "step": 320
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.8567,
+      "loss": 0.857,
       "step": 330
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.8761,
+      "loss": 0.8763,
       "step": 340
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.8834,
+      "loss": 0.8833,
       "step": 350
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.853,
+      "loss": 0.852,
       "step": 360
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.8389,
+      "loss": 0.8391,
       "step": 370
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.8699,
+      "loss": 0.8704,
       "step": 380
     },
     {
@@ -315,174 +315,174 @@
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.8786,
+      "loss": 0.8787,
       "step": 400
     },
     {
       "epoch": 0.15,
-      "eval_loss": 0.8281272053718567,
-      "eval_runtime": 107.4092,
-      "eval_samples_per_second": 9.31,
-      "eval_steps_per_second": 4.655,
+      "eval_loss": 0.8278040289878845,
+      "eval_runtime": 77.6481,
+      "eval_samples_per_second": 12.879,
+      "eval_steps_per_second": 6.439,
       "step": 400
     },
     {
       "epoch": 0.15,
-      "mmlu_eval_accuracy": 0.4990323912767152,
+      "mmlu_eval_accuracy": 0.5031268654890525,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
-      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5862068965517241,
-      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_biology": 0.375,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
       "mmlu_eval_accuracy_college_medicine": 0.5,
-      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
-      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
       "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
       "mmlu_eval_accuracy_econometrics": 0.25,
-      "mmlu_eval_accuracy_electrical_engineering": 0.4375,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
       "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
-      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
       "mmlu_eval_accuracy_global_facts": 0.4,
       "mmlu_eval_accuracy_high_school_biology": 0.46875,
       "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
       "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
       "mmlu_eval_accuracy_high_school_psychology": 0.85,
-      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
-      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
       "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
       "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
-      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
       "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.8,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
       "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
-      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
       "mmlu_eval_accuracy_moral_scenarios": 0.23,
-      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
       "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
       "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
-      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
-      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
-      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
       "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.4444444444444444,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.361526542248365,
+      "mmlu_loss": 1.3460789318184316,
       "step": 400
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.8911,
+      "loss": 0.8902,
       "step": 410
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.877,
+      "loss": 0.8767,
       "step": 420
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.8606,
+      "loss": 0.861,
       "step": 430
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.799,
+      "loss": 0.7996,
       "step": 440
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.8649,
+      "loss": 0.8643,
       "step": 450
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.8427,
+      "loss": 0.8435,
       "step": 460
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.8459,
+      "loss": 0.8466,
       "step": 470
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.8817,
+      "loss": 0.882,
       "step": 480
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.839,
+      "loss": 0.8397,
       "step": 490
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.8864,
+      "loss": 0.8861,
       "step": 500
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.88,
+      "loss": 0.8807,
       "step": 510
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.8291,
+      "loss": 0.8295,
       "step": 520
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.8677,
+      "loss": 0.8679,
       "step": 530
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.8545,
+      "loss": 0.8553,
       "step": 540
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.8127,
+      "loss": 0.8132,
       "step": 550
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.8645,
+      "loss": 0.8646,
       "step": 560
     },
     {
@@ -494,39 +494,39 @@
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.8417,
+      "loss": 0.8419,
       "step": 580
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8492,
+      "loss": 0.8491,
       "step": 590
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8885,
+      "loss": 0.889,
       "step": 600
     },
     {
       "epoch": 0.23,
-      "eval_loss": 0.8218681812286377,
-      "eval_runtime": 104.5102,
-      "eval_samples_per_second": 9.568,
-      "eval_steps_per_second": 4.784,
+      "eval_loss": 0.8218608498573303,
+      "eval_runtime": 77.4866,
+      "eval_samples_per_second": 12.905,
+      "eval_steps_per_second": 6.453,
       "step": 600
     },
     {
       "epoch": 0.23,
-      "mmlu_eval_accuracy": 0.49538660501170306,
+      "mmlu_eval_accuracy": 0.4952568072182041,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5862068965517241,
       "mmlu_eval_accuracy_college_biology": 0.4375,
-      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
       "mmlu_eval_accuracy_college_mathematics": 0.36363636363636365,
       "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
@@ -534,211 +534,211 @@
       "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
       "mmlu_eval_accuracy_conceptual_physics": 0.5,
       "mmlu_eval_accuracy_econometrics": 0.25,
-      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.1875,
       "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
       "mmlu_eval_accuracy_global_facts": 0.6,
-      "mmlu_eval_accuracy_high_school_biology": 0.5,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_biology": 0.53125,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
       "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
-      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.058823529411764705,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
       "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
-      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
       "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
-      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
       "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
-      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.19,
       "mmlu_eval_accuracy_nutrition": 0.5757575757575758,
-      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
-      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
-      "mmlu_eval_accuracy_professional_accounting": 0.25806451612903225,
-      "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
-      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
-      "mmlu_eval_accuracy_virology": 0.4444444444444444,
-      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.287742399554651,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.631578947368421,
+      "mmlu_loss": 1.2524041397686727,
       "step": 600
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8324,
+      "loss": 0.8316,
       "step": 610
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.8416,
+      "loss": 0.8418,
       "step": 620
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.8874,
+      "loss": 0.8881,
       "step": 630
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.8202,
+      "loss": 0.8215,
       "step": 640
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.8681,
+      "loss": 0.8684,
       "step": 650
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.8615,
+      "loss": 0.8619,
       "step": 660
     },
     {
       "epoch": 0.26,
       "learning_rate": 0.0002,
-      "loss": 0.829,
+      "loss": 0.8295,
       "step": 670
     },
     {
       "epoch": 0.26,
       "learning_rate": 0.0002,
-      "loss": 0.8438,
+      "loss": 0.8435,
       "step": 680
     },
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.8663,
+      "loss": 0.8655,
       "step": 690
     },
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.8501,
+      "loss": 0.8509,
       "step": 700
     },
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.8092,
+      "loss": 0.8085,
       "step": 710
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.8161,
+      "loss": 0.8175,
       "step": 720
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.8422,
+      "loss": 0.8427,
       "step": 730
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.8295,
+      "loss": 0.8289,
       "step": 740
     },
     {
       "epoch": 0.29,
       "learning_rate": 0.0002,
-      "loss": 0.8512,
+      "loss": 0.8511,
       "step": 750
     },
     {
       "epoch": 0.29,
       "learning_rate": 0.0002,
-      "loss": 0.8887,
+      "loss": 0.8882,
       "step": 760
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.8391,
+      "loss": 0.8395,
       "step": 770
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.7827,
+      "loss": 0.783,
       "step": 780
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.8772,
+      "loss": 0.8773,
       "step": 790
     },
     {
       "epoch": 0.31,
       "learning_rate": 0.0002,
-      "loss": 0.8513,
+      "loss": 0.8512,
       "step": 800
     },
     {
       "epoch": 0.31,
-      "eval_loss": 0.8209574222564697,
-      "eval_runtime": 104.3549,
-      "eval_samples_per_second": 9.583,
-      "eval_steps_per_second": 4.791,
+      "eval_loss": 0.8205058574676514,
+      "eval_runtime": 77.7581,
+      "eval_samples_per_second": 12.86,
+      "eval_steps_per_second": 6.43,
       "step": 800
     },
     {
       "epoch": 0.31,
-      "mmlu_eval_accuracy": 0.5051291111486879,
+      "mmlu_eval_accuracy": 0.49867192488087825,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
-      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5862068965517241,
       "mmlu_eval_accuracy_college_biology": 0.375,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
-      "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
       "mmlu_eval_accuracy_college_medicine": 0.5,
-      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
       "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
-      "mmlu_eval_accuracy_conceptual_physics": 0.5,
+      "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
       "mmlu_eval_accuracy_econometrics": 0.25,
-      "mmlu_eval_accuracy_electrical_engineering": 0.375,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
-      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
-      "mmlu_eval_accuracy_global_facts": 0.5,
-      "mmlu_eval_accuracy_high_school_biology": 0.46875,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_electrical_engineering": 0.4375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.3,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.37209302325581395,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
-      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
       "mmlu_eval_accuracy_high_school_psychology": 0.85,
       "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
       "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
@@ -751,24 +751,24 @@
       "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.8,
-      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
-      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
-      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.7093023255813954,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
       "mmlu_eval_accuracy_moral_scenarios": 0.24,
-      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
-      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
       "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
       "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
-      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
+      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
       "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
-      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.2605031216346876,
+      "mmlu_loss": 1.2541572008836364,
       "step": 800
     }
   ],

checkpoint-800/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68347c5c0bc67308405eaf249524c55462e787942d6d74b927105955b931cf02
+oid sha256:376563167fb75ef06e6c8600f9d035897ffea2363fc5fd45e427161c33ec50e3
 size 5819