Training in progress, step 5000

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8fee93695005d29240581b4f22c7d3ac9e402bd90ba9e650ce7db94c62b3367
+oid sha256:a83d3eaa152867b88924924d7763ddebb13854bf9346689e76408b7626436e63
 size 319977229

checkpoint-2800/adapter_model/adapter_model/README.md

@@ -1,32 +0,0 @@
----
-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
-- 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
-### Framework versions
-
-- PEFT 0.4.0
-
-- PEFT 0.4.0

checkpoint-2800/adapter_model/adapter_model/adapter_config.json

@@ -1,26 +0,0 @@
-{
-  "auto_mapping": null,
-  "base_model_name_or_path": "pankajmathur/orca_mini_v3_7b",
-  "bias": "none",
-  "fan_in_fan_out": false,
-  "inference_mode": true,
-  "init_lora_weights": true,
-  "layers_pattern": null,
-  "layers_to_transform": null,
-  "lora_alpha": 16.0,
-  "lora_dropout": 0.1,
-  "modules_to_save": null,
-  "peft_type": "LORA",
-  "r": 64,
-  "revision": null,
-  "target_modules": [
-    "gate_proj",
-    "k_proj",
-    "q_proj",
-    "v_proj",
-    "down_proj",
-    "up_proj",
-    "o_proj"
-  ],
-  "task_type": "CAUSAL_LM"
-}

checkpoint-2800/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f5fe610592da86c5b22c82ee4549d332c4058f9d8a8e7cd6ecf23a6025dc893b
-size 319977229

checkpoint-4000/adapter_model/adapter_model/README.md

@@ -37,6 +37,17 @@ The following `bitsandbytes` quantization config was used during training:
 - 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
+- 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
@@ -52,5 +63,6 @@ The following `bitsandbytes` quantization config was used during training:
 - PEFT 0.4.0
 - PEFT 0.4.0
 - PEFT 0.4.0
+- PEFT 0.4.0
 
 - PEFT 0.4.0

checkpoint-4000/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a96e67afa4c89ea922e0099c1f77ba98b299091089765ecd19bc6843ee8173d8
+oid sha256:d8fee93695005d29240581b4f22c7d3ac9e402bd90ba9e650ce7db94c62b3367
 size 319977229

{checkpoint-2800 → checkpoint-5000}/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e42a35eefa0ccd8b3e1e3b3b4c29c23eb966903891a8816b4c4c071e5a7a3687
+oid sha256:a83d3eaa152867b88924924d7763ddebb13854bf9346689e76408b7626436e63
 size 319977229

{checkpoint-2800 → checkpoint-5000}/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:492549fe6a72699a29d283336de02511c8bfd46a667acd852b8be25c229c787f
+oid sha256:1f5f63359cd6096b7ba04ecf7319ee927abf3fa788c2469bd00e300fa35ba53c
 size 1279539973

{checkpoint-2800 → checkpoint-5000}/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2ce4b7197b8a62d30c74121d1b1586b9f644b79ff862cd58e1f188e8c03eb3f
+oid sha256:0fe6ce5a2546a15dfe943e1c6fdc04bd04f4e2e8227d6e7890a4aa9a78d01c9e
 size 14511

{checkpoint-2800 → checkpoint-5000}/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79c08a7148c570c3d4c87a735e3acef3af52d256eb019f3a3f3feeab8a656949
+oid sha256:48d79510a393105d03ace4b25923bb16766b9918be883f120389bf31e783c069
 size 627

{checkpoint-2800 → checkpoint-5000}/trainer_state.json

@@ -1,8 +1,8 @@
 {
-  "best_metric": 0.7013482451438904,
-  "best_model_checkpoint": "experts/expert-9/checkpoint-2800",
-  "epoch": 1.3476115990855493,
-  "global_step": 2800,
+  "best_metric": 0.6935309767723083,
+  "best_model_checkpoint": "experts/expert-9/checkpoint-4000",
+  "epoch": 2.406449284081338,
+  "global_step": 5000,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -2680,11 +2680,2112 @@
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
       "mmlu_loss": 1.2200810726800102,
       "step": 2800
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.0002,
+      "loss": 0.6398,
+      "step": 2810
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002,
+      "loss": 0.6551,
+      "step": 2820
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002,
+      "loss": 0.6306,
+      "step": 2830
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002,
+      "loss": 0.5925,
+      "step": 2840
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002,
+      "loss": 0.6522,
+      "step": 2850
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002,
+      "loss": 0.6365,
+      "step": 2860
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002,
+      "loss": 0.6312,
+      "step": 2870
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0002,
+      "loss": 0.6168,
+      "step": 2880
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0002,
+      "loss": 0.6917,
+      "step": 2890
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002,
+      "loss": 0.6114,
+      "step": 2900
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002,
+      "loss": 0.6612,
+      "step": 2910
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002,
+      "loss": 0.5669,
+      "step": 2920
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002,
+      "loss": 0.6473,
+      "step": 2930
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002,
+      "loss": 0.6317,
+      "step": 2940
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002,
+      "loss": 0.6611,
+      "step": 2950
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002,
+      "loss": 0.6156,
+      "step": 2960
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.0002,
+      "loss": 0.6514,
+      "step": 2970
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.0002,
+      "loss": 0.6368,
+      "step": 2980
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002,
+      "loss": 0.6121,
+      "step": 2990
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002,
+      "loss": 0.6391,
+      "step": 3000
+    },
+    {
+      "epoch": 1.44,
+      "eval_loss": 0.7019394636154175,
+      "eval_runtime": 96.5558,
+      "eval_samples_per_second": 10.357,
+      "eval_steps_per_second": 5.178,
+      "step": 3000
+    },
+    {
+      "epoch": 1.44,
+      "mmlu_eval_accuracy": 0.4853465973418973,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_econometrics": 0.08333333333333333,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "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": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3103448275862069,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.34615384615384615,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.5263157894736842,
+      "mmlu_eval_accuracy_moral_scenarios": 0.29,
+      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.1684038531297802,
+      "step": 3000
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002,
+      "loss": 0.6525,
+      "step": 3010
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002,
+      "loss": 0.6265,
+      "step": 3020
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002,
+      "loss": 0.5704,
+      "step": 3030
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002,
+      "loss": 0.6634,
+      "step": 3040
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002,
+      "loss": 0.6899,
+      "step": 3050
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002,
+      "loss": 0.6469,
+      "step": 3060
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002,
+      "loss": 0.6063,
+      "step": 3070
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002,
+      "loss": 0.617,
+      "step": 3080
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.0002,
+      "loss": 0.5917,
+      "step": 3090
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.0002,
+      "loss": 0.6616,
+      "step": 3100
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002,
+      "loss": 0.6719,
+      "step": 3110
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002,
+      "loss": 0.6399,
+      "step": 3120
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002,
+      "loss": 0.6856,
+      "step": 3130
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002,
+      "loss": 0.5705,
+      "step": 3140
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002,
+      "loss": 0.6613,
+      "step": 3150
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002,
+      "loss": 0.6274,
+      "step": 3160
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002,
+      "loss": 0.6658,
+      "step": 3170
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002,
+      "loss": 0.6352,
+      "step": 3180
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.6471,
+      "step": 3190
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.632,
+      "step": 3200
+    },
+    {
+      "epoch": 1.54,
+      "eval_loss": 0.7007029056549072,
+      "eval_runtime": 96.7273,
+      "eval_samples_per_second": 10.338,
+      "eval_steps_per_second": 5.169,
+      "step": 3200
+    },
+    {
+      "epoch": 1.54,
+      "mmlu_eval_accuracy": 0.5016841158660759,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "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": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "mmlu_eval_accuracy_econometrics": 0.3333333333333333,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "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_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
+      "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": 0.2727272727272727,
+      "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": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.18,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3588235294117647,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.3333333333333333,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.0545151401359791,
+      "step": 3200
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.6669,
+      "step": 3210
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002,
+      "loss": 0.631,
+      "step": 3220
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002,
+      "loss": 0.6271,
+      "step": 3230
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.6661,
+      "step": 3240
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.5809,
+      "step": 3250
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.5527,
+      "step": 3260
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.6357,
+      "step": 3270
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002,
+      "loss": 0.6659,
+      "step": 3280
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002,
+      "loss": 0.6133,
+      "step": 3290
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.6268,
+      "step": 3300
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.614,
+      "step": 3310
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.6681,
+      "step": 3320
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.6475,
+      "step": 3330
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.6277,
+      "step": 3340
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.6327,
+      "step": 3350
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5997,
+      "step": 3360
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.6252,
+      "step": 3370
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.6061,
+      "step": 3380
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5826,
+      "step": 3390
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.6737,
+      "step": 3400
+    },
+    {
+      "epoch": 1.64,
+      "eval_loss": 0.7008338570594788,
+      "eval_runtime": 96.566,
+      "eval_samples_per_second": 10.356,
+      "eval_steps_per_second": 5.178,
+      "step": 3400
+    },
+    {
+      "epoch": 1.64,
+      "mmlu_eval_accuracy": 0.5059968735916802,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "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": 0.46153846153846156,
+      "mmlu_eval_accuracy_econometrics": 0.4166666666666667,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634,
+      "mmlu_eval_accuracy_formal_logic": 0.35714285714285715,
+      "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_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": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "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": 0.4782608695652174,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.84,
+      "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_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.35294117647058826,
+      "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": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.3888888888888889,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.1631466560357544,
+      "step": 3400
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.6199,
+      "step": 3410
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.6898,
+      "step": 3420
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.6113,
+      "step": 3430
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.6403,
+      "step": 3440
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5836,
+      "step": 3450
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5926,
+      "step": 3460
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.6327,
+      "step": 3470
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.6409,
+      "step": 3480
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.676,
+      "step": 3490
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.6035,
+      "step": 3500
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.5925,
+      "step": 3510
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.5939,
+      "step": 3520
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.6332,
+      "step": 3530
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.6247,
+      "step": 3540
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.6634,
+      "step": 3550
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.6296,
+      "step": 3560
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002,
+      "loss": 0.698,
+      "step": 3570
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002,
+      "loss": 0.6047,
+      "step": 3580
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.6075,
+      "step": 3590
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.6569,
+      "step": 3600
+    },
+    {
+      "epoch": 1.73,
+      "eval_loss": 0.6949359774589539,
+      "eval_runtime": 96.648,
+      "eval_samples_per_second": 10.347,
+      "eval_steps_per_second": 5.173,
+      "step": 3600
+    },
+    {
+      "epoch": 1.73,
+      "mmlu_eval_accuracy": 0.4935483523034825,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.36363636363636365,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "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_medicine": 0.5454545454545454,
+      "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": 0.3333333333333333,
+      "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": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "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": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "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": 0.84,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3588235294117647,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.3333333333333333,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.0990863862026765,
+      "step": 3600
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.6306,
+      "step": 3610
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.6212,
+      "step": 3620
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.6428,
+      "step": 3630
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.6247,
+      "step": 3640
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5999,
+      "step": 3650
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.6153,
+      "step": 3660
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.6349,
+      "step": 3670
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.6589,
+      "step": 3680
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.6412,
+      "step": 3690
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.6172,
+      "step": 3700
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.6902,
+      "step": 3710
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.6463,
+      "step": 3720
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5946,
+      "step": 3730
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.6402,
+      "step": 3740
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.6855,
+      "step": 3750
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.6514,
+      "step": 3760
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.6223,
+      "step": 3770
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.6242,
+      "step": 3780
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.661,
+      "step": 3790
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.6234,
+      "step": 3800
+    },
+    {
+      "epoch": 1.83,
+      "eval_loss": 0.6974416971206665,
+      "eval_runtime": 96.5494,
+      "eval_samples_per_second": 10.357,
+      "eval_steps_per_second": 5.179,
+      "step": 3800
+    },
+    {
+      "epoch": 1.83,
+      "mmlu_eval_accuracy": 0.5132864378444664,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "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_medicine": 0.5,
+      "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": 0.3333333333333333,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "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_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "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": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "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_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
+      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.8421052631578947,
+      "mmlu_loss": 1.197790414094925,
+      "step": 3800
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5663,
+      "step": 3810
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.6275,
+      "step": 3820
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.6492,
+      "step": 3830
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.6502,
+      "step": 3840
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.6415,
+      "step": 3850
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.6535,
+      "step": 3860
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.6023,
+      "step": 3870
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002,
+      "loss": 0.6052,
+      "step": 3880
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002,
+      "loss": 0.6309,
+      "step": 3890
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.5763,
+      "step": 3900
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.6168,
+      "step": 3910
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.6134,
+      "step": 3920
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.6515,
+      "step": 3930
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.6632,
+      "step": 3940
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.6673,
+      "step": 3950
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.6382,
+      "step": 3960
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.6102,
+      "step": 3970
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.6235,
+      "step": 3980
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.6015,
+      "step": 3990
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.6513,
+      "step": 4000
+    },
+    {
+      "epoch": 1.93,
+      "eval_loss": 0.6935309767723083,
+      "eval_runtime": 96.5168,
+      "eval_samples_per_second": 10.361,
+      "eval_steps_per_second": 5.18,
+      "step": 4000
+    },
+    {
+      "epoch": 1.93,
+      "mmlu_eval_accuracy": 0.5048661718395844,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "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_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_econometrics": 0.4166666666666667,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "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_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "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": 0.7391304347826086,
+      "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": 0.6111111111111112,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.4117647058823529,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.8421052631578947,
+      "mmlu_loss": 1.0965017748719408,
+      "step": 4000
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.5982,
+      "step": 4010
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.6548,
+      "step": 4020
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.6693,
+      "step": 4030
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5996,
+      "step": 4040
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.647,
+      "step": 4050
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.6158,
+      "step": 4060
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002,
+      "loss": 0.6132,
+      "step": 4070
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002,
+      "loss": 0.6379,
+      "step": 4080
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0002,
+      "loss": 0.6187,
+      "step": 4090
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0002,
+      "loss": 0.628,
+      "step": 4100
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0002,
+      "loss": 0.6404,
+      "step": 4110
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0002,
+      "loss": 0.6648,
+      "step": 4120
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0002,
+      "loss": 0.6049,
+      "step": 4130
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0002,
+      "loss": 0.5937,
+      "step": 4140
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0002,
+      "loss": 0.635,
+      "step": 4150
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0002,
+      "loss": 0.5441,
+      "step": 4160
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002,
+      "loss": 0.5747,
+      "step": 4170
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002,
+      "loss": 0.493,
+      "step": 4180
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0002,
+      "loss": 0.5512,
+      "step": 4190
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0002,
+      "loss": 0.5196,
+      "step": 4200
+    },
+    {
+      "epoch": 2.02,
+      "eval_loss": 0.7101156115531921,
+      "eval_runtime": 96.495,
+      "eval_samples_per_second": 10.363,
+      "eval_steps_per_second": 5.182,
+      "step": 4200
+    },
+    {
+      "epoch": 2.02,
+      "mmlu_eval_accuracy": 0.5156976828735957,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "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": 0.4090909090909091,
+      "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": 0.4166666666666667,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "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": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.5116279069767442,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.7833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6111111111111112,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.84,
+      "mmlu_eval_accuracy_medical_genetics": 1.0,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.6296296296296297,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
+      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.8421052631578947,
+      "mmlu_loss": 1.172741430371455,
+      "step": 4200
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0002,
+      "loss": 0.4644,
+      "step": 4210
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0002,
+      "loss": 0.5029,
+      "step": 4220
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002,
+      "loss": 0.5515,
+      "step": 4230
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002,
+      "loss": 0.56,
+      "step": 4240
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002,
+      "loss": 0.5708,
+      "step": 4250
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002,
+      "loss": 0.5187,
+      "step": 4260
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002,
+      "loss": 0.5577,
+      "step": 4270
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002,
+      "loss": 0.4832,
+      "step": 4280
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002,
+      "loss": 0.531,
+      "step": 4290
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0002,
+      "loss": 0.5103,
+      "step": 4300
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0002,
+      "loss": 0.5374,
+      "step": 4310
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002,
+      "loss": 0.5239,
+      "step": 4320
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002,
+      "loss": 0.5476,
+      "step": 4330
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002,
+      "loss": 0.4858,
+      "step": 4340
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002,
+      "loss": 0.5068,
+      "step": 4350
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002,
+      "loss": 0.5509,
+      "step": 4360
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002,
+      "loss": 0.5908,
+      "step": 4370
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0002,
+      "loss": 0.5936,
+      "step": 4380
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0002,
+      "loss": 0.4865,
+      "step": 4390
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0002,
+      "loss": 0.5701,
+      "step": 4400
+    },
+    {
+      "epoch": 2.12,
+      "eval_loss": 0.708640992641449,
+      "eval_runtime": 96.4973,
+      "eval_samples_per_second": 10.363,
+      "eval_steps_per_second": 5.181,
+      "step": 4400
+    },
+    {
+      "epoch": 2.12,
+      "mmlu_eval_accuracy": 0.5114226272922625,
+      "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "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": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.3333333333333333,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.59375,
+      "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_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "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,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.88,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.35294117647058826,
+      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "mmlu_eval_accuracy_professional_psychology": 0.4492753623188406,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "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": 0.7894736842105263,
+      "mmlu_loss": 1.116310212479092,
+      "step": 4400
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0002,
+      "loss": 0.5463,
+      "step": 4410
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0002,
+      "loss": 0.491,
+      "step": 4420
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0002,
+      "loss": 0.5565,
+      "step": 4430
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002,
+      "loss": 0.4916,
+      "step": 4440
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002,
+      "loss": 0.54,
+      "step": 4450
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0002,
+      "loss": 0.5678,
+      "step": 4460
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0002,
+      "loss": 0.5326,
+      "step": 4470
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002,
+      "loss": 0.5559,
+      "step": 4480
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002,
+      "loss": 0.5155,
+      "step": 4490
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002,
+      "loss": 0.512,
+      "step": 4500
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002,
+      "loss": 0.5508,
+      "step": 4510
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0002,
+      "loss": 0.5643,
+      "step": 4520
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0002,
+      "loss": 0.5063,
+      "step": 4530
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002,
+      "loss": 0.6117,
+      "step": 4540
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002,
+      "loss": 0.5134,
+      "step": 4550
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002,
+      "loss": 0.5298,
+      "step": 4560
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0002,
+      "loss": 0.4778,
+      "step": 4570
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0002,
+      "loss": 0.507,
+      "step": 4580
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0002,
+      "loss": 0.5872,
+      "step": 4590
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0002,
+      "loss": 0.5005,
+      "step": 4600
+    },
+    {
+      "epoch": 2.21,
+      "eval_loss": 0.7128604054450989,
+      "eval_runtime": 96.6906,
+      "eval_samples_per_second": 10.342,
+      "eval_steps_per_second": 5.171,
+      "step": 4600
+    },
+    {
+      "epoch": 2.21,
+      "mmlu_eval_accuracy": 0.5042506599448242,
+      "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_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.43902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.3,
+      "mmlu_eval_accuracy_high_school_biology": 0.5,
+      "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_geography": 0.8636363636363636,
+      "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": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8,
+      "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": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6111111111111112,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
+      "mmlu_eval_accuracy_management": 0.5454545454545454,
+      "mmlu_eval_accuracy_marketing": 0.88,
+      "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": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.0643581137177838,
+      "step": 4600
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0002,
+      "loss": 0.5507,
+      "step": 4610
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0002,
+      "loss": 0.536,
+      "step": 4620
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0002,
+      "loss": 0.5877,
+      "step": 4630
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0002,
+      "loss": 0.5081,
+      "step": 4640
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002,
+      "loss": 0.5627,
+      "step": 4650
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002,
+      "loss": 0.5184,
+      "step": 4660
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0002,
+      "loss": 0.5258,
+      "step": 4670
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0002,
+      "loss": 0.5677,
+      "step": 4680
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002,
+      "loss": 0.5301,
+      "step": 4690
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002,
+      "loss": 0.573,
+      "step": 4700
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0002,
+      "loss": 0.5683,
+      "step": 4710
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0002,
+      "loss": 0.5356,
+      "step": 4720
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002,
+      "loss": 0.5307,
+      "step": 4730
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002,
+      "loss": 0.5409,
+      "step": 4740
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002,
+      "loss": 0.5407,
+      "step": 4750
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002,
+      "loss": 0.5777,
+      "step": 4760
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002,
+      "loss": 0.5876,
+      "step": 4770
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002,
+      "loss": 0.5316,
+      "step": 4780
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002,
+      "loss": 0.5129,
+      "step": 4790
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002,
+      "loss": 0.5671,
+      "step": 4800
+    },
+    {
+      "epoch": 2.31,
+      "eval_loss": 0.717536211013794,
+      "eval_runtime": 96.6151,
+      "eval_samples_per_second": 10.35,
+      "eval_steps_per_second": 5.175,
+      "step": 4800
+    },
+    {
+      "epoch": 2.31,
+      "mmlu_eval_accuracy": 0.48998732953958063,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "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_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.59375,
+      "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": 0.7727272727272727,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.5116279069767442,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8,
+      "mmlu_eval_accuracy_high_school_statistics": 0.2608695652173913,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7727272727272727,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6538461538461539,
+      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
+      "mmlu_eval_accuracy_management": 0.5454545454545454,
+      "mmlu_eval_accuracy_marketing": 0.84,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.4117647058823529,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.4492753623188406,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.5555555555555556,
+      "mmlu_eval_accuracy_sociology": 0.5909090909090909,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.631578947368421,
+      "mmlu_loss": 1.0062876213779648,
+      "step": 4800
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002,
+      "loss": 0.5615,
+      "step": 4810
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002,
+      "loss": 0.5557,
+      "step": 4820
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002,
+      "loss": 0.4921,
+      "step": 4830
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002,
+      "loss": 0.576,
+      "step": 4840
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002,
+      "loss": 0.5185,
+      "step": 4850
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0002,
+      "loss": 0.5836,
+      "step": 4860
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0002,
+      "loss": 0.5826,
+      "step": 4870
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002,
+      "loss": 0.5376,
+      "step": 4880
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002,
+      "loss": 0.5577,
+      "step": 4890
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002,
+      "loss": 0.5041,
+      "step": 4900
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002,
+      "loss": 0.5831,
+      "step": 4910
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002,
+      "loss": 0.574,
+      "step": 4920
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002,
+      "loss": 0.5621,
+      "step": 4930
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0002,
+      "loss": 0.5102,
+      "step": 4940
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0002,
+      "loss": 0.503,
+      "step": 4950
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002,
+      "loss": 0.5425,
+      "step": 4960
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002,
+      "loss": 0.5057,
+      "step": 4970
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0002,
+      "loss": 0.5723,
+      "step": 4980
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0002,
+      "loss": 0.496,
+      "step": 4990
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0002,
+      "loss": 0.5237,
+      "step": 5000
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.7113077640533447,
+      "eval_runtime": 96.6394,
+      "eval_samples_per_second": 10.348,
+      "eval_steps_per_second": 5.174,
+      "step": 5000
+    },
+    {
+      "epoch": 2.41,
+      "mmlu_eval_accuracy": 0.49230804644830667,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "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": 0.45454545454545453,
+      "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": 0.3333333333333333,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.3,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.4090909090909091,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.75,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "mmlu_eval_accuracy_human_sexuality": 0.4166666666666667,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.5,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.5454545454545454,
+      "mmlu_eval_accuracy_marketing": 0.84,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.5454545454545454,
+      "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.5925925925925926,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.8421052631578947,
+      "mmlu_loss": 1.1302715398701924,
+      "step": 5000
     }
   ],
   "max_steps": 10000,
   "num_train_epochs": 5,
-  "total_flos": 7.415619108188652e+17,
+  "total_flos": 1.3195894528270664e+18,
   "trial_name": null,
   "trial_params": null
 }