Training in progress, step 1800

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84689ae7bdc6e21d613632f903cef740863cea93dd88e14fbfe22be5b9034705
+oid sha256:4fb6f31d5816137072037b26d50671bc3bc0aaf5e7ac6e09047216937637be3f
 size 319977229

checkpoint-1600/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-1600/adapter_model/adapter_model/adapter_config.json

@@ -14,10 +14,10 @@
   "r": 64,
   "revision": null,
   "target_modules": [
-    "o_proj",
+    "v_proj",
     "q_proj",
+    "o_proj",
     "down_proj",
-    "v_proj",
     "k_proj",
     "gate_proj",
     "up_proj"

checkpoint-1600/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29cb0bd6303c8413bf8494050c8eab64182be4dc35d0a7615406ec7b8d28ca87
+oid sha256:84689ae7bdc6e21d613632f903cef740863cea93dd88e14fbfe22be5b9034705
 size 319977229

checkpoint-1800/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-1800/adapter_config.json

@@ -14,10 +14,10 @@
   "r": 64,
   "revision": null,
   "target_modules": [
-    "o_proj",
+    "v_proj",
     "q_proj",
+    "o_proj",
     "down_proj",
-    "v_proj",
     "k_proj",
     "gate_proj",
     "up_proj"

checkpoint-1800/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efdc2a905cceb1a9c3bfe194c1b9885e9dfdbbe4afd906a34a44a8c1628ee9e5
+oid sha256:4fb6f31d5816137072037b26d50671bc3bc0aaf5e7ac6e09047216937637be3f
 size 319977229

checkpoint-1800/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:157c09f13bad518035ba714e0096d902cbe548561408ec73523c67a40aff9a32
+oid sha256:a2b75eccf13617d469d573a6744e8b348a8a5ade36c495ff4c094c1ff0839c7e
 size 1279539973

checkpoint-1800/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1495e11e7e5238ca19f18e3523d226fd180e23a4a31be918cd8dfe87181933a
+oid sha256:51229b7760247c605d7dc63e0ac9782db249cdf6489371fe2ce671b39e202459
 size 14511

checkpoint-1800/trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.9326393008232117,
+  "best_metric": 0.9326978921890259,
   "best_model_checkpoint": "experts/expert-14/checkpoint-1800",
   "epoch": 0.5536332179930796,
   "global_step": 1800,
@@ -10,13 +10,13 @@
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 1.0634,
+      "loss": 1.064,
       "step": 10
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 1.0357,
+      "loss": 1.0358,
       "step": 20
     },
     {
@@ -28,43 +28,43 @@
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.9858,
+      "loss": 0.9861,
       "step": 40
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 1.0149,
+      "loss": 1.0148,
       "step": 50
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 1.0292,
+      "loss": 1.0293,
       "step": 60
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.9947,
+      "loss": 0.9945,
       "step": 70
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 1.0068,
+      "loss": 1.0069,
       "step": 80
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 1.015,
+      "loss": 1.0147,
       "step": 90
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 1.0055,
+      "loss": 1.0056,
       "step": 100
     },
     {
@@ -76,37 +76,37 @@
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 1.0218,
+      "loss": 1.0214,
       "step": 120
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.9645,
+      "loss": 0.9642,
       "step": 130
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 1.0059,
+      "loss": 1.0051,
       "step": 140
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 1.0041,
+      "loss": 1.004,
       "step": 150
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.9626,
+      "loss": 0.9631,
       "step": 160
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.9682,
+      "loss": 0.9674,
       "step": 170
     },
     {
@@ -118,74 +118,74 @@
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 1.0029,
+      "loss": 1.0023,
       "step": 190
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.9873,
+      "loss": 0.9879,
       "step": 200
     },
     {
       "epoch": 0.06,
-      "eval_loss": 0.9700731039047241,
-      "eval_runtime": 90.2222,
-      "eval_samples_per_second": 11.084,
-      "eval_steps_per_second": 5.542,
+      "eval_loss": 0.970292329788208,
+      "eval_runtime": 67.4255,
+      "eval_samples_per_second": 14.831,
+      "eval_steps_per_second": 7.416,
       "step": 200
     },
     {
       "epoch": 0.06,
-      "mmlu_eval_accuracy": 0.4896808303512203,
+      "mmlu_eval_accuracy": 0.49145352624106475,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
-      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
       "mmlu_eval_accuracy_college_biology": 0.3125,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
       "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
       "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
-      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
       "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
       "mmlu_eval_accuracy_econometrics": 0.25,
       "mmlu_eval_accuracy_electrical_engineering": 0.3125,
       "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
-      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
-      "mmlu_eval_accuracy_global_facts": 0.6,
-      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "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": 0.6111111111111112,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
       "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": 0.6363636363636364,
-      "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
-      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.6086956521739131,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
       "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
       "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
       "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": 0.6511627906976745,
-      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.23,
-      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
-      "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
       "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
-      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
       "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
       "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
       "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
@@ -195,7 +195,7 @@
       "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
       "mmlu_eval_accuracy_virology": 0.4444444444444444,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.6262522289242507,
+      "mmlu_loss": 1.6139244634862357,
       "step": 200
     },
     {
@@ -207,13 +207,13 @@
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 1.0338,
+      "loss": 1.0343,
       "step": 220
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 1.0196,
+      "loss": 1.0197,
       "step": 230
     },
     {
@@ -231,134 +231,134 @@
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.9735,
+      "loss": 0.9737,
       "step": 260
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.9661,
+      "loss": 0.9663,
       "step": 270
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 1.0139,
+      "loss": 1.0132,
       "step": 280
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.993,
+      "loss": 0.9933,
       "step": 290
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.9777,
+      "loss": 0.9779,
       "step": 300
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 1.0059,
+      "loss": 1.0064,
       "step": 310
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.9822,
+      "loss": 0.982,
       "step": 320
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.9382,
+      "loss": 0.9386,
       "step": 330
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 1.0069,
+      "loss": 1.0075,
       "step": 340
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 1.0168,
+      "loss": 1.0162,
       "step": 350
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.9999,
+      "loss": 0.9997,
       "step": 360
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.9736,
+      "loss": 0.9735,
       "step": 370
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.9869,
+      "loss": 0.9864,
       "step": 380
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.9386,
+      "loss": 0.9384,
       "step": 390
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.9746,
+      "loss": 0.974,
       "step": 400
     },
     {
       "epoch": 0.12,
-      "eval_loss": 0.9594073295593262,
-      "eval_runtime": 90.2621,
-      "eval_samples_per_second": 11.079,
-      "eval_steps_per_second": 5.539,
+      "eval_loss": 0.9592995643615723,
+      "eval_runtime": 67.7732,
+      "eval_samples_per_second": 14.755,
+      "eval_steps_per_second": 7.378,
       "step": 400
     },
     {
       "epoch": 0.12,
-      "mmlu_eval_accuracy": 0.5010359222325911,
-      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy": 0.5052135585190854,
+      "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
       "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
-      "mmlu_eval_accuracy_astronomy": 0.375,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
       "mmlu_eval_accuracy_clinical_knowledge": 0.6206896551724138,
-      "mmlu_eval_accuracy_college_biology": 0.25,
+      "mmlu_eval_accuracy_college_biology": 0.3125,
       "mmlu_eval_accuracy_college_chemistry": 0.375,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
-      "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
       "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
       "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
-      "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
-      "mmlu_eval_accuracy_conceptual_physics": 0.5,
+      "mmlu_eval_accuracy_computer_security": 0.8181818181818182,
+      "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
       "mmlu_eval_accuracy_econometrics": 0.25,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
       "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
       "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
       "mmlu_eval_accuracy_global_facts": 0.4,
-      "mmlu_eval_accuracy_high_school_biology": 0.5,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_biology": 0.53125,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "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": 0.6666666666666666,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023,
       "mmlu_eval_accuracy_high_school_mathematics": 0.13793103448275862,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
       "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
+      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
       "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
       "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
@@ -370,23 +370,23 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
-      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.21,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
       "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
-      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_philosophy": 0.5,
       "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
       "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
-      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
       "mmlu_eval_accuracy_professional_medicine": 0.4838709677419355,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
-      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_sociology": 0.8181818181818182,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.3888888888888889,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.4568967482284216,
+      "mmlu_loss": 1.4457430965595395,
       "step": 400
     },
     {
@@ -398,73 +398,73 @@
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 1.0163,
+      "loss": 1.0161,
       "step": 420
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.9342,
+      "loss": 0.9338,
       "step": 430
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.9329,
+      "loss": 0.9332,
       "step": 440
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 1.0435,
+      "loss": 1.0441,
       "step": 450
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.9783,
+      "loss": 0.9784,
       "step": 460
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.9395,
+      "loss": 0.9396,
       "step": 470
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.9889,
+      "loss": 0.9896,
       "step": 480
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 1.0053,
+      "loss": 1.0052,
       "step": 490
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.9455,
+      "loss": 0.9449,
       "step": 500
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 1.009,
+      "loss": 1.0091,
       "step": 510
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 1.0258,
+      "loss": 1.0261,
       "step": 520
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.9632,
+      "loss": 0.9628,
       "step": 530
     },
     {
@@ -476,84 +476,84 @@
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.9982,
+      "loss": 0.9979,
       "step": 550
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.9807,
+      "loss": 0.9803,
       "step": 560
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 1.0157,
+      "loss": 1.016,
       "step": 570
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.9795,
+      "loss": 0.9794,
       "step": 580
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.9964,
+      "loss": 0.9952,
       "step": 590
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.9949,
+      "loss": 0.9937,
       "step": 600
     },
     {
       "epoch": 0.18,
-      "eval_loss": 0.9541609287261963,
-      "eval_runtime": 90.288,
-      "eval_samples_per_second": 11.076,
-      "eval_steps_per_second": 5.538,
+      "eval_loss": 0.9536917805671692,
+      "eval_runtime": 67.3333,
+      "eval_samples_per_second": 14.851,
+      "eval_steps_per_second": 7.426,
       "step": 600
     },
     {
       "epoch": 0.18,
-      "mmlu_eval_accuracy": 0.504817046060058,
-      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy": 0.5083307909266076,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
-      "mmlu_eval_accuracy_astronomy": 0.375,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
-      "mmlu_eval_accuracy_college_biology": 0.5625,
-      "mmlu_eval_accuracy_college_chemistry": 0.125,
-      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
       "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
       "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
       "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
-      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
       "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
-      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
-      "mmlu_eval_accuracy_global_facts": 0.3,
+      "mmlu_eval_accuracy_global_facts": 0.4,
       "mmlu_eval_accuracy_high_school_biology": 0.34375,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
       "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": 0.6190476190476191,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
-      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
       "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
-      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
       "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
-      "mmlu_eval_accuracy_human_sexuality": 0.5,
+      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
       "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
@@ -562,64 +562,64 @@
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 1.0,
       "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
-      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
-      "mmlu_eval_accuracy_moral_scenarios": 0.24,
-      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
       "mmlu_eval_accuracy_philosophy": 0.4117647058823529,
       "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
-      "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_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.5,
       "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
       "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
-      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_virology": 0.5,
       "mmlu_eval_accuracy_world_religions": 0.8421052631578947,
-      "mmlu_loss": 1.398965796664863,
+      "mmlu_loss": 1.416625464398307,
       "step": 600
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 1.0161,
+      "loss": 1.016,
       "step": 610
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.9693,
+      "loss": 0.9692,
       "step": 620
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.955,
+      "loss": 0.9542,
       "step": 630
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.9782,
+      "loss": 0.9776,
       "step": 640
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.941,
+      "loss": 0.9416,
       "step": 650
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.9831,
+      "loss": 0.983,
       "step": 660
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.9776,
+      "loss": 0.9767,
       "step": 670
     },
     {
@@ -631,43 +631,43 @@
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.9678,
+      "loss": 0.9682,
       "step": 690
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.9399,
+      "loss": 0.9387,
       "step": 700
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.993,
+      "loss": 0.9928,
       "step": 710
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.9686,
+      "loss": 0.9683,
       "step": 720
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 1.0153,
+      "loss": 1.0156,
       "step": 730
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.9634,
+      "loss": 0.9632,
       "step": 740
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.9307,
+      "loss": 0.9312,
       "step": 750
     },
     {
@@ -679,68 +679,68 @@
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.9925,
+      "loss": 0.9929,
       "step": 770
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.9762,
+      "loss": 0.9763,
       "step": 780
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.9963,
+      "loss": 0.9965,
       "step": 790
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.9464,
+      "loss": 0.9468,
       "step": 800
     },
     {
       "epoch": 0.25,
-      "eval_loss": 0.9473406076431274,
-      "eval_runtime": 90.2453,
-      "eval_samples_per_second": 11.081,
-      "eval_steps_per_second": 5.54,
+      "eval_loss": 0.9470635056495667,
+      "eval_runtime": 67.3149,
+      "eval_samples_per_second": 14.856,
+      "eval_steps_per_second": 7.428,
       "step": 800
     },
     {
       "epoch": 0.25,
-      "mmlu_eval_accuracy": 0.5006352396162203,
+      "mmlu_eval_accuracy": 0.5018069141697037,
       "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
-      "mmlu_eval_accuracy_astronomy": 0.5,
-      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
       "mmlu_eval_accuracy_college_biology": 0.4375,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
-      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
-      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "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": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.3125,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
       "mmlu_eval_accuracy_global_facts": 0.5,
-      "mmlu_eval_accuracy_high_school_biology": 0.40625,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_biology": 0.375,
+      "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_european_history": 0.6111111111111112,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
       "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
-      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
-      "mmlu_eval_accuracy_high_school_psychology": 0.85,
-      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
       "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
       "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
@@ -748,57 +748,57 @@
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
       "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
-      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
-      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
       "mmlu_eval_accuracy_moral_scenarios": 0.24,
       "mmlu_eval_accuracy_nutrition": 0.696969696969697,
-      "mmlu_eval_accuracy_philosophy": 0.38235294117647056,
-      "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
+      "mmlu_eval_accuracy_philosophy": 0.4117647058823529,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
       "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
-      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
-      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
-      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
-      "mmlu_eval_accuracy_public_relations": 0.5,
-      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
       "mmlu_eval_accuracy_virology": 0.4444444444444444,
       "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
-      "mmlu_loss": 1.4785823159834113,
+      "mmlu_loss": 1.4953326111362746,
       "step": 800
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.9627,
+      "loss": 0.9628,
       "step": 810
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.9633,
+      "loss": 0.9632,
       "step": 820
     },
     {
       "epoch": 0.26,
       "learning_rate": 0.0002,
-      "loss": 0.9925,
+      "loss": 0.993,
       "step": 830
     },
     {
       "epoch": 0.26,
       "learning_rate": 0.0002,
-      "loss": 0.9914,
+      "loss": 0.9913,
       "step": 840
     },
     {
       "epoch": 0.26,
       "learning_rate": 0.0002,
-      "loss": 0.9461,
+      "loss": 0.9459,
       "step": 850
     },
     {
@@ -810,37 +810,37 @@
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.9754,
+      "loss": 0.9746,
       "step": 870
     },
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.9824,
+      "loss": 0.9829,
       "step": 880
     },
     {
       "epoch": 0.27,
       "learning_rate": 0.0002,
-      "loss": 0.9718,
+      "loss": 0.9707,
       "step": 890
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.9684,
+      "loss": 0.9685,
       "step": 900
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.9534,
+      "loss": 0.9542,
       "step": 910
     },
     {
       "epoch": 0.28,
       "learning_rate": 0.0002,
-      "loss": 0.9796,
+      "loss": 0.9804,
       "step": 920
     },
     {
@@ -852,89 +852,89 @@
     {
       "epoch": 0.29,
       "learning_rate": 0.0002,
-      "loss": 1.0022,
+      "loss": 1.0025,
       "step": 940
     },
     {
       "epoch": 0.29,
       "learning_rate": 0.0002,
-      "loss": 0.9577,
+      "loss": 0.9576,
       "step": 950
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.9768,
+      "loss": 0.9772,
       "step": 960
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.9689,
+      "loss": 0.9691,
       "step": 970
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.983,
+      "loss": 0.9833,
       "step": 980
     },
     {
       "epoch": 0.3,
       "learning_rate": 0.0002,
-      "loss": 0.9745,
+      "loss": 0.9741,
       "step": 990
     },
     {
       "epoch": 0.31,
       "learning_rate": 0.0002,
-      "loss": 0.9356,
+      "loss": 0.9368,
       "step": 1000
     },
     {
       "epoch": 0.31,
-      "eval_loss": 0.9430285692214966,
-      "eval_runtime": 90.214,
-      "eval_samples_per_second": 11.085,
-      "eval_steps_per_second": 5.542,
+      "eval_loss": 0.9433629512786865,
+      "eval_runtime": 67.2667,
+      "eval_samples_per_second": 14.866,
+      "eval_steps_per_second": 7.433,
       "step": 1000
     },
     {
       "epoch": 0.31,
-      "mmlu_eval_accuracy": 0.49482155317463744,
+      "mmlu_eval_accuracy": 0.4854609382018465,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
-      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_anatomy": 0.5,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
-      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.375,
       "mmlu_eval_accuracy_college_chemistry": 0.125,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
       "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
-      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
       "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
       "mmlu_eval_accuracy_econometrics": 0.08333333333333333,
-      "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
       "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
-      "mmlu_eval_accuracy_global_facts": 0.6,
-      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.375,
       "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.5,
-      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6666666666666666,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
       "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
       "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
       "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
-      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "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,
@@ -943,189 +943,189 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
       "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
       "mmlu_eval_accuracy_moral_scenarios": 0.24,
       "mmlu_eval_accuracy_nutrition": 0.5454545454545454,
-      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_philosophy": 0.5,
       "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
       "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
-      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
-      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
       "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
-      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
       "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
-      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.631578947368421,
-      "mmlu_loss": 1.448764714964376,
+      "mmlu_loss": 1.484623718355094,
       "step": 1000
     },
     {
       "epoch": 0.31,
       "learning_rate": 0.0002,
-      "loss": 1.0236,
+      "loss": 1.0237,
       "step": 1010
     },
     {
       "epoch": 0.31,
       "learning_rate": 0.0002,
-      "loss": 0.9802,
+      "loss": 0.98,
       "step": 1020
     },
     {
       "epoch": 0.32,
       "learning_rate": 0.0002,
-      "loss": 0.9495,
+      "loss": 0.9486,
       "step": 1030
     },
     {
       "epoch": 0.32,
       "learning_rate": 0.0002,
-      "loss": 1.0232,
+      "loss": 1.0235,
       "step": 1040
     },
     {
       "epoch": 0.32,
       "learning_rate": 0.0002,
-      "loss": 0.9533,
+      "loss": 0.9547,
       "step": 1050
     },
     {
       "epoch": 0.33,
       "learning_rate": 0.0002,
-      "loss": 0.9756,
+      "loss": 0.9743,
       "step": 1060
     },
     {
       "epoch": 0.33,
       "learning_rate": 0.0002,
-      "loss": 0.9643,
+      "loss": 0.964,
       "step": 1070
     },
     {
       "epoch": 0.33,
       "learning_rate": 0.0002,
-      "loss": 0.9752,
+      "loss": 0.9744,
       "step": 1080
     },
     {
       "epoch": 0.34,
       "learning_rate": 0.0002,
-      "loss": 0.9924,
+      "loss": 0.9927,
       "step": 1090
     },
     {
       "epoch": 0.34,
       "learning_rate": 0.0002,
-      "loss": 0.9781,
+      "loss": 0.9776,
       "step": 1100
     },
     {
       "epoch": 0.34,
       "learning_rate": 0.0002,
-      "loss": 0.977,
+      "loss": 0.9771,
       "step": 1110
     },
     {
       "epoch": 0.34,
       "learning_rate": 0.0002,
-      "loss": 0.9588,
+      "loss": 0.9579,
       "step": 1120
     },
     {
       "epoch": 0.35,
       "learning_rate": 0.0002,
-      "loss": 0.9758,
+      "loss": 0.9754,
       "step": 1130
     },
     {
       "epoch": 0.35,
       "learning_rate": 0.0002,
-      "loss": 0.9855,
+      "loss": 0.9856,
       "step": 1140
     },
     {
       "epoch": 0.35,
       "learning_rate": 0.0002,
-      "loss": 0.9493,
+      "loss": 0.9486,
       "step": 1150
     },
     {
       "epoch": 0.36,
       "learning_rate": 0.0002,
-      "loss": 0.997,
+      "loss": 0.9968,
       "step": 1160
     },
     {
       "epoch": 0.36,
       "learning_rate": 0.0002,
-      "loss": 0.9486,
+      "loss": 0.9482,
       "step": 1170
     },
     {
       "epoch": 0.36,
       "learning_rate": 0.0002,
-      "loss": 0.9671,
+      "loss": 0.9667,
       "step": 1180
     },
     {
       "epoch": 0.37,
       "learning_rate": 0.0002,
-      "loss": 0.9584,
+      "loss": 0.957,
       "step": 1190
     },
     {
       "epoch": 0.37,
       "learning_rate": 0.0002,
-      "loss": 0.9492,
+      "loss": 0.9487,
       "step": 1200
     },
     {
       "epoch": 0.37,
-      "eval_loss": 0.9415192604064941,
-      "eval_runtime": 90.2902,
-      "eval_samples_per_second": 11.075,
-      "eval_steps_per_second": 5.538,
+      "eval_loss": 0.9412919878959656,
+      "eval_runtime": 67.3263,
+      "eval_samples_per_second": 14.853,
+      "eval_steps_per_second": 7.427,
       "step": 1200
     },
     {
       "epoch": 0.37,
-      "mmlu_eval_accuracy": 0.5037954074669568,
-      "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
-      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy": 0.5030946594890945,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.3125,
       "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": 0.125,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_chemistry": 0.375,
       "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
-      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
-      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
-      "mmlu_eval_accuracy_computer_security": 0.8181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5454545454545454,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
       "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
       "mmlu_eval_accuracy_econometrics": 0.25,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
       "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
-      "mmlu_eval_accuracy_global_facts": 0.5,
-      "mmlu_eval_accuracy_high_school_biology": 0.5625,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.18181818181818182,
+      "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_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.5,
       "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
       "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
-      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
       "mmlu_eval_accuracy_high_school_statistics": 0.5217391304347826,
       "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
-      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
-      "mmlu_eval_accuracy_human_aging": 0.6086956521739131,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.5652173913043478,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
       "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
@@ -1134,188 +1134,188 @@
       "mmlu_eval_accuracy_management": 0.7272727272727273,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
-      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
-      "mmlu_eval_accuracy_moral_disputes": 0.5263157894736842,
-      "mmlu_eval_accuracy_moral_scenarios": 0.22,
-      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
-      "mmlu_eval_accuracy_philosophy": 0.5,
-      "mmlu_eval_accuracy_prehistory": 0.6285714285714286,
-      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
-      "mmlu_eval_accuracy_professional_law": 0.38235294117647056,
-      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
-      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.21,
+      "mmlu_eval_accuracy_nutrition": 0.5757575757575758,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.6,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.36470588235294116,
+      "mmlu_eval_accuracy_professional_medicine": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
-      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
-      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "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": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.631578947368421,
-      "mmlu_loss": 1.3555281070053733,
+      "mmlu_loss": 1.380024835506245,
       "step": 1200
     },
     {
       "epoch": 0.37,
       "learning_rate": 0.0002,
-      "loss": 0.9424,
+      "loss": 0.9426,
       "step": 1210
     },
     {
       "epoch": 0.38,
       "learning_rate": 0.0002,
-      "loss": 1.0029,
+      "loss": 1.0035,
       "step": 1220
     },
     {
       "epoch": 0.38,
       "learning_rate": 0.0002,
-      "loss": 0.9852,
+      "loss": 0.9848,
       "step": 1230
     },
     {
       "epoch": 0.38,
       "learning_rate": 0.0002,
-      "loss": 0.9422,
+      "loss": 0.9415,
       "step": 1240
     },
     {
       "epoch": 0.38,
       "learning_rate": 0.0002,
-      "loss": 1.0081,
+      "loss": 1.008,
       "step": 1250
     },
     {
       "epoch": 0.39,
       "learning_rate": 0.0002,
-      "loss": 0.9712,
+      "loss": 0.9711,
       "step": 1260
     },
     {
       "epoch": 0.39,
       "learning_rate": 0.0002,
-      "loss": 1.0181,
+      "loss": 1.0182,
       "step": 1270
     },
     {
       "epoch": 0.39,
       "learning_rate": 0.0002,
-      "loss": 0.9494,
+      "loss": 0.9504,
       "step": 1280
     },
     {
       "epoch": 0.4,
       "learning_rate": 0.0002,
-      "loss": 1.0123,
+      "loss": 1.0109,
       "step": 1290
     },
     {
       "epoch": 0.4,
       "learning_rate": 0.0002,
-      "loss": 0.9592,
+      "loss": 0.9589,
       "step": 1300
     },
     {
       "epoch": 0.4,
       "learning_rate": 0.0002,
-      "loss": 0.9293,
+      "loss": 0.9291,
       "step": 1310
     },
     {
       "epoch": 0.41,
       "learning_rate": 0.0002,
-      "loss": 0.9751,
+      "loss": 0.9749,
       "step": 1320
     },
     {
       "epoch": 0.41,
       "learning_rate": 0.0002,
-      "loss": 0.9361,
+      "loss": 0.9352,
       "step": 1330
     },
     {
       "epoch": 0.41,
       "learning_rate": 0.0002,
-      "loss": 0.9757,
+      "loss": 0.9762,
       "step": 1340
     },
     {
       "epoch": 0.42,
       "learning_rate": 0.0002,
-      "loss": 0.9431,
+      "loss": 0.9409,
       "step": 1350
     },
     {
       "epoch": 0.42,
       "learning_rate": 0.0002,
-      "loss": 0.9881,
+      "loss": 0.9874,
       "step": 1360
     },
     {
       "epoch": 0.42,
       "learning_rate": 0.0002,
-      "loss": 0.9374,
+      "loss": 0.9369,
       "step": 1370
     },
     {
       "epoch": 0.42,
       "learning_rate": 0.0002,
-      "loss": 0.974,
+      "loss": 0.9741,
       "step": 1380
     },
     {
       "epoch": 0.43,
       "learning_rate": 0.0002,
-      "loss": 0.9666,
+      "loss": 0.9667,
       "step": 1390
     },
     {
       "epoch": 0.43,
       "learning_rate": 0.0002,
-      "loss": 0.9408,
+      "loss": 0.9416,
       "step": 1400
     },
     {
       "epoch": 0.43,
-      "eval_loss": 0.9398794174194336,
-      "eval_runtime": 90.5066,
-      "eval_samples_per_second": 11.049,
-      "eval_steps_per_second": 5.524,
+      "eval_loss": 0.9397566318511963,
+      "eval_runtime": 68.0239,
+      "eval_samples_per_second": 14.701,
+      "eval_steps_per_second": 7.35,
       "step": 1400
     },
     {
       "epoch": 0.43,
-      "mmlu_eval_accuracy": 0.49614480814477097,
-      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy": 0.4934082826400804,
+      "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.4375,
-      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
       "mmlu_eval_accuracy_college_biology": 0.4375,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
       "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
-      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
-      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
-      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
-      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
+      "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "mmlu_eval_accuracy_econometrics": 0.08333333333333333,
       "mmlu_eval_accuracy_electrical_engineering": 0.1875,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
-      "mmlu_eval_accuracy_global_facts": 0.5,
-      "mmlu_eval_accuracy_high_school_biology": 0.5,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.5,
+      "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.5769230769230769,
-      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3103448275862069,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
       "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
-      "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6153846153846154,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
@@ -1325,95 +1325,95 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.88,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
-      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
-      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
       "mmlu_eval_accuracy_nutrition": 0.5757575757575758,
-      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
-      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
       "mmlu_eval_accuracy_professional_accounting": 0.45161290322580644,
-      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
-      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
-      "mmlu_eval_accuracy_professional_psychology": 0.5217391304347826,
-      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
       "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
-      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5,
-      "mmlu_eval_accuracy_world_religions": 0.631578947368421,
-      "mmlu_loss": 1.3774613221382659,
+      "mmlu_eval_accuracy_world_religions": 0.5789473684210527,
+      "mmlu_loss": 1.4450731767686789,
       "step": 1400
     },
     {
       "epoch": 0.43,
       "learning_rate": 0.0002,
-      "loss": 0.9399,
+      "loss": 0.9401,
       "step": 1410
     },
     {
       "epoch": 0.44,
       "learning_rate": 0.0002,
-      "loss": 1.0147,
+      "loss": 1.0137,
       "step": 1420
     },
     {
       "epoch": 0.44,
       "learning_rate": 0.0002,
-      "loss": 1.004,
+      "loss": 1.0035,
       "step": 1430
     },
     {
       "epoch": 0.44,
       "learning_rate": 0.0002,
-      "loss": 0.9569,
+      "loss": 0.9578,
       "step": 1440
     },
     {
       "epoch": 0.45,
       "learning_rate": 0.0002,
-      "loss": 0.9759,
+      "loss": 0.9751,
       "step": 1450
     },
     {
       "epoch": 0.45,
       "learning_rate": 0.0002,
-      "loss": 0.9548,
+      "loss": 0.9547,
       "step": 1460
     },
     {
       "epoch": 0.45,
       "learning_rate": 0.0002,
-      "loss": 0.9839,
+      "loss": 0.9843,
       "step": 1470
     },
     {
       "epoch": 0.46,
       "learning_rate": 0.0002,
-      "loss": 0.9769,
+      "loss": 0.9775,
       "step": 1480
     },
     {
       "epoch": 0.46,
       "learning_rate": 0.0002,
-      "loss": 0.9701,
+      "loss": 0.9698,
       "step": 1490
     },
     {
       "epoch": 0.46,
       "learning_rate": 0.0002,
-      "loss": 0.9742,
+      "loss": 0.9747,
       "step": 1500
     },
     {
       "epoch": 0.46,
       "learning_rate": 0.0002,
-      "loss": 0.9708,
+      "loss": 0.9703,
       "step": 1510
     },
     {
       "epoch": 0.47,
       "learning_rate": 0.0002,
-      "loss": 0.9609,
+      "loss": 0.9618,
       "step": 1520
     },
     {
@@ -1425,7 +1425,7 @@
     {
       "epoch": 0.47,
       "learning_rate": 0.0002,
-      "loss": 0.9267,
+      "loss": 0.9273,
       "step": 1540
     },
     {
@@ -1437,77 +1437,77 @@
     {
       "epoch": 0.48,
       "learning_rate": 0.0002,
-      "loss": 1.0096,
+      "loss": 1.0102,
       "step": 1560
     },
     {
       "epoch": 0.48,
       "learning_rate": 0.0002,
-      "loss": 0.9567,
+      "loss": 0.9564,
       "step": 1570
     },
     {
       "epoch": 0.49,
       "learning_rate": 0.0002,
-      "loss": 0.9563,
+      "loss": 0.9566,
       "step": 1580
     },
     {
       "epoch": 0.49,
       "learning_rate": 0.0002,
-      "loss": 0.9979,
+      "loss": 0.9991,
       "step": 1590
     },
     {
       "epoch": 0.49,
       "learning_rate": 0.0002,
-      "loss": 0.9615,
+      "loss": 0.9605,
       "step": 1600
     },
     {
       "epoch": 0.49,
-      "eval_loss": 0.9336153268814087,
-      "eval_runtime": 90.3488,
-      "eval_samples_per_second": 11.068,
-      "eval_steps_per_second": 5.534,
+      "eval_loss": 0.9335065484046936,
+      "eval_runtime": 67.7146,
+      "eval_samples_per_second": 14.768,
+      "eval_steps_per_second": 7.384,
       "step": 1600
     },
     {
       "epoch": 0.49,
-      "mmlu_eval_accuracy": 0.4851638399139786,
+      "mmlu_eval_accuracy": 0.4889036635942836,
       "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
       "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
-      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_astronomy": 0.375,
       "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
       "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
       "mmlu_eval_accuracy_college_biology": 0.3125,
       "mmlu_eval_accuracy_college_chemistry": 0.125,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
-      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
       "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
       "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
       "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
       "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
       "mmlu_eval_accuracy_econometrics": 0.08333333333333333,
       "mmlu_eval_accuracy_electrical_engineering": 0.1875,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
       "mmlu_eval_accuracy_global_facts": 0.5,
-      "mmlu_eval_accuracy_high_school_biology": 0.5,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
       "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
       "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
-      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
       "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
       "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
-      "mmlu_eval_accuracy_human_aging": 0.6086956521739131,
+      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
       "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
       "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
@@ -1516,77 +1516,77 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
-      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.23,
-      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.5757575757575758,
       "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
       "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
-      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
-      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
+      "mmlu_eval_accuracy_professional_accounting": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
       "mmlu_eval_accuracy_professional_medicine": 0.4838709677419355,
-      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_professional_psychology": 0.5217391304347826,
       "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
       "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
-      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.2181525310165888,
+      "mmlu_loss": 1.2396197211649025,
       "step": 1600
     },
     {
       "epoch": 0.5,
       "learning_rate": 0.0002,
-      "loss": 0.9779,
+      "loss": 0.9788,
       "step": 1610
     },
     {
       "epoch": 0.5,
       "learning_rate": 0.0002,
-      "loss": 0.9303,
+      "loss": 0.9306,
       "step": 1620
     },
     {
       "epoch": 0.5,
       "learning_rate": 0.0002,
-      "loss": 0.9676,
+      "loss": 0.9668,
       "step": 1630
     },
     {
       "epoch": 0.5,
       "learning_rate": 0.0002,
-      "loss": 0.9568,
+      "loss": 0.9565,
       "step": 1640
     },
     {
       "epoch": 0.51,
       "learning_rate": 0.0002,
-      "loss": 0.9949,
+      "loss": 0.9957,
       "step": 1650
     },
     {
       "epoch": 0.51,
       "learning_rate": 0.0002,
-      "loss": 0.9144,
+      "loss": 0.9135,
       "step": 1660
     },
     {
       "epoch": 0.51,
       "learning_rate": 0.0002,
-      "loss": 0.9454,
+      "loss": 0.9461,
       "step": 1670
     },
     {
       "epoch": 0.52,
       "learning_rate": 0.0002,
-      "loss": 0.996,
+      "loss": 0.9965,
       "step": 1680
     },
     {
       "epoch": 0.52,
       "learning_rate": 0.0002,
-      "loss": 0.9615,
+      "loss": 0.962,
       "step": 1690
     },
     {
@@ -1598,132 +1598,132 @@
     {
       "epoch": 0.53,
       "learning_rate": 0.0002,
-      "loss": 0.898,
+      "loss": 0.897,
       "step": 1710
     },
     {
       "epoch": 0.53,
       "learning_rate": 0.0002,
-      "loss": 0.9632,
+      "loss": 0.9637,
       "step": 1720
     },
     {
       "epoch": 0.53,
       "learning_rate": 0.0002,
-      "loss": 0.9629,
+      "loss": 0.9635,
       "step": 1730
     },
     {
       "epoch": 0.54,
       "learning_rate": 0.0002,
-      "loss": 0.9506,
+      "loss": 0.9519,
       "step": 1740
     },
     {
       "epoch": 0.54,
       "learning_rate": 0.0002,
-      "loss": 0.9463,
+      "loss": 0.9453,
       "step": 1750
     },
     {
       "epoch": 0.54,
       "learning_rate": 0.0002,
-      "loss": 0.9733,
+      "loss": 0.9741,
       "step": 1760
     },
     {
       "epoch": 0.54,
       "learning_rate": 0.0002,
-      "loss": 0.9663,
+      "loss": 0.9669,
       "step": 1770
     },
     {
       "epoch": 0.55,
       "learning_rate": 0.0002,
-      "loss": 0.9783,
+      "loss": 0.979,
       "step": 1780
     },
     {
       "epoch": 0.55,
       "learning_rate": 0.0002,
-      "loss": 0.9531,
+      "loss": 0.9536,
       "step": 1790
     },
     {
       "epoch": 0.55,
       "learning_rate": 0.0002,
-      "loss": 0.9662,
+      "loss": 0.9669,
       "step": 1800
     },
     {
       "epoch": 0.55,
-      "eval_loss": 0.9326393008232117,
-      "eval_runtime": 90.6614,
-      "eval_samples_per_second": 11.03,
-      "eval_steps_per_second": 5.515,
+      "eval_loss": 0.9326978921890259,
+      "eval_runtime": 67.7454,
+      "eval_samples_per_second": 14.761,
+      "eval_steps_per_second": 7.381,
       "step": 1800
     },
     {
       "epoch": 0.55,
-      "mmlu_eval_accuracy": 0.4956669607823638,
-      "mmlu_eval_accuracy_abstract_algebra": 0.5454545454545454,
-      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy": 0.4992790250348134,
+      "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.5,
-      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
-      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.375,
       "mmlu_eval_accuracy_college_chemistry": 0.125,
       "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
       "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
-      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
-      "mmlu_eval_accuracy_college_physics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
       "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
       "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_elementary_mathematics": 0.36585365853658536,
       "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
-      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_global_facts": 0.3,
       "mmlu_eval_accuracy_high_school_biology": 0.40625,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
-      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
       "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
-      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
       "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
       "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
-      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
-      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
-      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
       "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": 0.7209302325581395,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
       "mmlu_eval_accuracy_moral_scenarios": 0.23,
       "mmlu_eval_accuracy_nutrition": 0.5757575757575758,
-      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
       "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
-      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_accounting": 0.5161290322580645,
       "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
-      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
-      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
-      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
       "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
-      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.631578947368421,
-      "mmlu_loss": 1.2750722245743629,
+      "mmlu_loss": 1.3108376859451398,
       "step": 1800
     }
   ],

checkpoint-1800/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:132dc0489795dd2e62e3dfed60f8fa6c2da9f513c3d6724e00171983c97e1c5b
+oid sha256:6c1f3b4e86a18f376afc61102db182028a2b670075c0fd60d0651751e2c0b01a
 size 5819