Training in progress, step 5000

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17e2827f6dc0ae2f350ed947449d4b71baf6c17995b288f32827e1971e15593f
+oid sha256:1fdae70ef9896e0f0a90df70b90b3c60a7ff549c38981c0ae6a764e7a0d53de6
 size 319977229

checkpoint-2800/adapter_model/adapter_model/README.md

@@ -1,20 +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
-### Framework versions
-
-
-- 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": [
-    "v_proj",
-    "q_proj",
-    "up_proj",
-    "gate_proj",
-    "down_proj",
-    "o_proj",
-    "k_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 sha256:577a2a6712ad2ba1e52441a36cc959c85157ae8f97a62963b6c6d566705011f7
-size 319977229

checkpoint-3400/adapter_model/adapter_model/README.md

@@ -70,6 +70,17 @@ 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
@@ -88,5 +99,6 @@ 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-3400/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9714af7e2292eb3a345542d7619207a4839b60032564280f1430af24ceb63d5d
+oid sha256:17e2827f6dc0ae2f350ed947449d4b71baf6c17995b288f32827e1971e15593f
 size 319977229

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

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:577a2a6712ad2ba1e52441a36cc959c85157ae8f97a62963b6c6d566705011f7
+oid sha256:1fdae70ef9896e0f0a90df70b90b3c60a7ff549c38981c0ae6a764e7a0d53de6
 size 319977229

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

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4a7a7c36966eaddcdf9895f8af7239f6e1501ae6162e2e8a673b2a71ef6fa73
+oid sha256:1abf78df81f0004329d3b25cb834be5814d118fc022d15bd3c61d631f46d1e9b
 size 1279539973

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

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f4a3320774614af968d7bec550edcfd4402bd34666d965c7c00ae67be372e51
+oid sha256:df7f04af940c559340d0ba0041d7e1733874b7e1ef99b10059f3d23c19e3196e
 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": 0.7184598445892334,
-  "best_model_checkpoint": "experts/expert-30/checkpoint-2800",
-  "epoch": 1.5581524763494712,
-  "global_step": 2800,
+  "best_metric": 0.7174572944641113,
+  "best_model_checkpoint": "experts/expert-30/checkpoint-3400",
+  "epoch": 2.782415136338342,
+  "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": 0.7368421052631579,
       "mmlu_loss": 1.6170813445945322,
       "step": 2800
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.6409,
+      "step": 2810
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.6334,
+      "step": 2820
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.5712,
+      "step": 2830
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002,
+      "loss": 0.7104,
+      "step": 2840
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.6489,
+      "step": 2850
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.679,
+      "step": 2860
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.6998,
+      "step": 2870
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.7515,
+      "step": 2880
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.6938,
+      "step": 2890
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.6724,
+      "step": 2900
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.708,
+      "step": 2910
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.761,
+      "step": 2920
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.7505,
+      "step": 2930
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.6852,
+      "step": 2940
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.7095,
+      "step": 2950
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.7221,
+      "step": 2960
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.8497,
+      "step": 2970
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.6602,
+      "step": 2980
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5926,
+      "step": 2990
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.6759,
+      "step": 3000
+    },
+    {
+      "epoch": 1.67,
+      "eval_loss": 0.7177545428276062,
+      "eval_runtime": 80.9037,
+      "eval_samples_per_second": 12.36,
+      "eval_steps_per_second": 6.18,
+      "step": 3000
+    },
+    {
+      "epoch": 1.67,
+      "mmlu_eval_accuracy": 0.5099074449177355,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.1875,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.10344827586206896,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.6153846153846154,
+      "mmlu_eval_accuracy_high_school_physics": 0.4117647058823529,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.7222222222222222,
+      "mmlu_eval_accuracy_machine_learning": 0.09090909090909091,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.36470588235294116,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.4057971014492754,
+      "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": 0.9090909090909091,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.5329999124568063,
+      "step": 3000
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.6581,
+      "step": 3010
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.6017,
+      "step": 3020
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.7058,
+      "step": 3030
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.6653,
+      "step": 3040
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.675,
+      "step": 3050
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.7569,
+      "step": 3060
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.6843,
+      "step": 3070
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.6791,
+      "step": 3080
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002,
+      "loss": 0.6962,
+      "step": 3090
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.6748,
+      "step": 3100
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.7165,
+      "step": 3110
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.6527,
+      "step": 3120
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.6898,
+      "step": 3130
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.6503,
+      "step": 3140
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.6628,
+      "step": 3150
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.6397,
+      "step": 3160
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.6749,
+      "step": 3170
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.7057,
+      "step": 3180
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.6982,
+      "step": 3190
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.7518,
+      "step": 3200
+    },
+    {
+      "epoch": 1.78,
+      "eval_loss": 0.7215366363525391,
+      "eval_runtime": 80.938,
+      "eval_samples_per_second": 12.355,
+      "eval_steps_per_second": 6.178,
+      "step": 3200
+    },
+    {
+      "epoch": 1.78,
+      "mmlu_eval_accuracy": 0.49686780674439057,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "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": 0.3125,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.13793103448275862,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5769230769230769,
+      "mmlu_eval_accuracy_high_school_physics": 0.4117647058823529,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
+      "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": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.09090909090909091,
+      "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_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "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": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.5558218805036719,
+      "step": 3200
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.5908,
+      "step": 3210
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.6995,
+      "step": 3220
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.6722,
+      "step": 3230
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.6415,
+      "step": 3240
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.6145,
+      "step": 3250
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.7593,
+      "step": 3260
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.6967,
+      "step": 3270
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.7044,
+      "step": 3280
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.6739,
+      "step": 3290
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.7549,
+      "step": 3300
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.6651,
+      "step": 3310
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.6304,
+      "step": 3320
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.697,
+      "step": 3330
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.7056,
+      "step": 3340
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.6866,
+      "step": 3350
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002,
+      "loss": 0.6961,
+      "step": 3360
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.6659,
+      "step": 3370
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.6562,
+      "step": 3380
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.7485,
+      "step": 3390
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.6648,
+      "step": 3400
+    },
+    {
+      "epoch": 1.89,
+      "eval_loss": 0.7174572944641113,
+      "eval_runtime": 81.0067,
+      "eval_samples_per_second": 12.345,
+      "eval_steps_per_second": 6.172,
+      "step": 3400
+    },
+    {
+      "epoch": 1.89,
+      "mmlu_eval_accuracy": 0.49318718415804697,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "mmlu_eval_accuracy_college_chemistry": 0.375,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
+      "mmlu_eval_accuracy_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_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": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "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": 0.2727272727272727,
+      "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": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "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": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.4525975414962744,
+      "step": 3400
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.6248,
+      "step": 3410
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.6542,
+      "step": 3420
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.6543,
+      "step": 3430
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.7032,
+      "step": 3440
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.6607,
+      "step": 3450
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.6923,
+      "step": 3460
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.6825,
+      "step": 3470
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.7021,
+      "step": 3480
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.7099,
+      "step": 3490
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.643,
+      "step": 3500
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.6298,
+      "step": 3510
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002,
+      "loss": 0.6354,
+      "step": 3520
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002,
+      "loss": 0.6184,
+      "step": 3530
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0002,
+      "loss": 0.6326,
+      "step": 3540
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0002,
+      "loss": 0.7001,
+      "step": 3550
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0002,
+      "loss": 0.7251,
+      "step": 3560
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0002,
+      "loss": 0.7081,
+      "step": 3570
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0002,
+      "loss": 0.6451,
+      "step": 3580
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0002,
+      "loss": 0.6656,
+      "step": 3590
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0002,
+      "loss": 0.5813,
+      "step": 3600
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.7238325476646423,
+      "eval_runtime": 80.9818,
+      "eval_samples_per_second": 12.348,
+      "eval_steps_per_second": 6.174,
+      "step": 3600
+    },
+    {
+      "epoch": 2.0,
+      "mmlu_eval_accuracy": 0.5006190661031859,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.7857142857142857,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "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": 0.34615384615384615,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "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": 0.40625,
+      "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_government_and_politics": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "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": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "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": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "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": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "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": 0.6842105263157895,
+      "mmlu_loss": 1.748088232219686,
+      "step": 3600
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002,
+      "loss": 0.5764,
+      "step": 3610
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002,
+      "loss": 0.5246,
+      "step": 3620
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0002,
+      "loss": 0.5854,
+      "step": 3630
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0002,
+      "loss": 0.5774,
+      "step": 3640
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0002,
+      "loss": 0.6143,
+      "step": 3650
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002,
+      "loss": 0.5396,
+      "step": 3660
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002,
+      "loss": 0.6007,
+      "step": 3670
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002,
+      "loss": 0.5761,
+      "step": 3680
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002,
+      "loss": 0.6143,
+      "step": 3690
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002,
+      "loss": 0.6141,
+      "step": 3700
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002,
+      "loss": 0.5369,
+      "step": 3710
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0002,
+      "loss": 0.5552,
+      "step": 3720
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002,
+      "loss": 0.5438,
+      "step": 3730
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002,
+      "loss": 0.5865,
+      "step": 3740
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002,
+      "loss": 0.5346,
+      "step": 3750
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002,
+      "loss": 0.6198,
+      "step": 3760
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002,
+      "loss": 0.5625,
+      "step": 3770
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002,
+      "loss": 0.5754,
+      "step": 3780
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0002,
+      "loss": 0.5826,
+      "step": 3790
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0002,
+      "loss": 0.6256,
+      "step": 3800
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.7389654517173767,
+      "eval_runtime": 80.8435,
+      "eval_samples_per_second": 12.37,
+      "eval_steps_per_second": 6.185,
+      "step": 3800
+    },
+    {
+      "epoch": 2.11,
+      "mmlu_eval_accuracy": 0.4874363553860279,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.3076923076923077,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "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": 0.8636363636363636,
+      "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": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "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": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "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": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.31,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.4057971014492754,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.8434810874200676,
+      "step": 3800
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0002,
+      "loss": 0.5493,
+      "step": 3810
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0002,
+      "loss": 0.6217,
+      "step": 3820
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0002,
+      "loss": 0.5834,
+      "step": 3830
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002,
+      "loss": 0.5555,
+      "step": 3840
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002,
+      "loss": 0.5916,
+      "step": 3850
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0002,
+      "loss": 0.5741,
+      "step": 3860
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0002,
+      "loss": 0.6221,
+      "step": 3870
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002,
+      "loss": 0.6192,
+      "step": 3880
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002,
+      "loss": 0.61,
+      "step": 3890
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002,
+      "loss": 0.6222,
+      "step": 3900
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0002,
+      "loss": 0.5747,
+      "step": 3910
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0002,
+      "loss": 0.6442,
+      "step": 3920
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002,
+      "loss": 0.6257,
+      "step": 3930
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002,
+      "loss": 0.6422,
+      "step": 3940
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0002,
+      "loss": 0.5695,
+      "step": 3950
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0002,
+      "loss": 0.6073,
+      "step": 3960
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0002,
+      "loss": 0.6101,
+      "step": 3970
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0002,
+      "loss": 0.5485,
+      "step": 3980
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0002,
+      "loss": 0.5537,
+      "step": 3990
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0002,
+      "loss": 0.6276,
+      "step": 4000
+    },
+    {
+      "epoch": 2.23,
+      "eval_loss": 0.7375339269638062,
+      "eval_runtime": 81.1838,
+      "eval_samples_per_second": 12.318,
+      "eval_steps_per_second": 6.159,
+      "step": 4000
+    },
+    {
+      "epoch": 2.23,
+      "mmlu_eval_accuracy": 0.4988898595282169,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "mmlu_eval_accuracy_college_biology": 0.5,
+      "mmlu_eval_accuracy_college_chemistry": 0.375,
+      "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_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "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": 0.14285714285714285,
+      "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": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.7222222222222222,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.27,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.6,
+      "mmlu_eval_accuracy_professional_accounting": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.5972761438658591,
+      "step": 4000
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0002,
+      "loss": 0.5624,
+      "step": 4010
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002,
+      "loss": 0.6697,
+      "step": 4020
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002,
+      "loss": 0.5763,
+      "step": 4030
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0002,
+      "loss": 0.5639,
+      "step": 4040
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0002,
+      "loss": 0.6299,
+      "step": 4050
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002,
+      "loss": 0.5273,
+      "step": 4060
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002,
+      "loss": 0.5685,
+      "step": 4070
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0002,
+      "loss": 0.6153,
+      "step": 4080
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002,
+      "loss": 0.6321,
+      "step": 4090
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002,
+      "loss": 0.633,
+      "step": 4100
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002,
+      "loss": 0.6263,
+      "step": 4110
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002,
+      "loss": 0.5602,
+      "step": 4120
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002,
+      "loss": 0.6217,
+      "step": 4130
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002,
+      "loss": 0.5929,
+      "step": 4140
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002,
+      "loss": 0.5451,
+      "step": 4150
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002,
+      "loss": 0.5998,
+      "step": 4160
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002,
+      "loss": 0.5706,
+      "step": 4170
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002,
+      "loss": 0.5579,
+      "step": 4180
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002,
+      "loss": 0.5578,
+      "step": 4190
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0002,
+      "loss": 0.6218,
+      "step": 4200
+    },
+    {
+      "epoch": 2.34,
+      "eval_loss": 0.7373414635658264,
+      "eval_runtime": 81.0652,
+      "eval_samples_per_second": 12.336,
+      "eval_steps_per_second": 6.168,
+      "step": 4200
+    },
+    {
+      "epoch": 2.34,
+      "mmlu_eval_accuracy": 0.4928563025913322,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "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": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "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": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "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": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.6086956521739131,
+      "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": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.28,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.4492753623188406,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.3888888888888889,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.717369836475457,
+      "step": 4200
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0002,
+      "loss": 0.5223,
+      "step": 4210
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002,
+      "loss": 0.5841,
+      "step": 4220
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002,
+      "loss": 0.6448,
+      "step": 4230
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002,
+      "loss": 0.5961,
+      "step": 4240
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002,
+      "loss": 0.6016,
+      "step": 4250
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002,
+      "loss": 0.5662,
+      "step": 4260
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0002,
+      "loss": 0.623,
+      "step": 4270
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0002,
+      "loss": 0.5799,
+      "step": 4280
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002,
+      "loss": 0.6123,
+      "step": 4290
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002,
+      "loss": 0.6131,
+      "step": 4300
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0002,
+      "loss": 0.6332,
+      "step": 4310
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0002,
+      "loss": 0.5694,
+      "step": 4320
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0002,
+      "loss": 0.6708,
+      "step": 4330
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0002,
+      "loss": 0.5414,
+      "step": 4340
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0002,
+      "loss": 0.6039,
+      "step": 4350
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0002,
+      "loss": 0.6201,
+      "step": 4360
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0002,
+      "loss": 0.6074,
+      "step": 4370
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0002,
+      "loss": 0.5907,
+      "step": 4380
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0002,
+      "loss": 0.5812,
+      "step": 4390
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0002,
+      "loss": 0.5774,
+      "step": 4400
+    },
+    {
+      "epoch": 2.45,
+      "eval_loss": 0.7349709272384644,
+      "eval_runtime": 81.2457,
+      "eval_samples_per_second": 12.308,
+      "eval_steps_per_second": 6.154,
+      "step": 4400
+    },
+    {
+      "epoch": 2.45,
+      "mmlu_eval_accuracy": 0.4971195440035668,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "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": 0.375,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "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": 0.3076923076923077,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "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": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_computer_science": 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": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023,
+      "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": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.21739130434782608,
+      "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": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.09090909090909091,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.72,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.29,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_law": 0.3764705882352941,
+      "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": 0.4444444444444444,
+      "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": 0.6842105263157895,
+      "mmlu_loss": 1.621530317995629,
+      "step": 4400
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0002,
+      "loss": 0.5676,
+      "step": 4410
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0002,
+      "loss": 0.6241,
+      "step": 4420
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0002,
+      "loss": 0.5596,
+      "step": 4430
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0002,
+      "loss": 0.6229,
+      "step": 4440
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0002,
+      "loss": 0.5909,
+      "step": 4450
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0002,
+      "loss": 0.6188,
+      "step": 4460
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0002,
+      "loss": 0.5837,
+      "step": 4470
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0002,
+      "loss": 0.6362,
+      "step": 4480
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0002,
+      "loss": 0.6077,
+      "step": 4490
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0002,
+      "loss": 0.6419,
+      "step": 4500
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0002,
+      "loss": 0.53,
+      "step": 4510
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0002,
+      "loss": 0.5789,
+      "step": 4520
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0002,
+      "loss": 0.5993,
+      "step": 4530
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0002,
+      "loss": 0.5547,
+      "step": 4540
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0002,
+      "loss": 0.5731,
+      "step": 4550
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0002,
+      "loss": 0.5699,
+      "step": 4560
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0002,
+      "loss": 0.6324,
+      "step": 4570
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0002,
+      "loss": 0.6007,
+      "step": 4580
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0002,
+      "loss": 0.5826,
+      "step": 4590
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0002,
+      "loss": 0.5998,
+      "step": 4600
+    },
+    {
+      "epoch": 2.56,
+      "eval_loss": 0.7342756390571594,
+      "eval_runtime": 81.2044,
+      "eval_samples_per_second": 12.315,
+      "eval_steps_per_second": 6.157,
+      "step": 4600
+    },
+    {
+      "epoch": 2.56,
+      "mmlu_eval_accuracy": 0.48445799959172553,
+      "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": 0.4375,
+      "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_computer_security": 0.45454545454545453,
+      "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": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
+      "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": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.2608695652173913,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.31,
+      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.35294117647058826,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.42028985507246375,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "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": 0.6842105263157895,
+      "mmlu_loss": 1.7353844049316163,
+      "step": 4600
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0002,
+      "loss": 0.5933,
+      "step": 4610
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0002,
+      "loss": 0.6088,
+      "step": 4620
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0002,
+      "loss": 0.5678,
+      "step": 4630
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0002,
+      "loss": 0.602,
+      "step": 4640
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0002,
+      "loss": 0.6084,
+      "step": 4650
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0002,
+      "loss": 0.6243,
+      "step": 4660
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0002,
+      "loss": 0.5954,
+      "step": 4670
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0002,
+      "loss": 0.6858,
+      "step": 4680
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0002,
+      "loss": 0.5645,
+      "step": 4690
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5635,
+      "step": 4700
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0002,
+      "loss": 0.6461,
+      "step": 4710
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0002,
+      "loss": 0.6703,
+      "step": 4720
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0002,
+      "loss": 0.6765,
+      "step": 4730
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0002,
+      "loss": 0.6172,
+      "step": 4740
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5559,
+      "step": 4750
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0002,
+      "loss": 0.5918,
+      "step": 4760
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0002,
+      "loss": 0.6178,
+      "step": 4770
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5901,
+      "step": 4780
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0002,
+      "loss": 0.6441,
+      "step": 4790
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5607,
+      "step": 4800
+    },
+    {
+      "epoch": 2.67,
+      "eval_loss": 0.7372880578041077,
+      "eval_runtime": 80.9653,
+      "eval_samples_per_second": 12.351,
+      "eval_steps_per_second": 6.175,
+      "step": 4800
+    },
+    {
+      "epoch": 2.67,
+      "mmlu_eval_accuracy": 0.49053452276115184,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "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": 0.36585365853658536,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.35,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.4057971014492754,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.5985829301820411,
+      "step": 4800
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5994,
+      "step": 4810
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0002,
+      "loss": 0.6289,
+      "step": 4820
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0002,
+      "loss": 0.6103,
+      "step": 4830
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0002,
+      "loss": 0.566,
+      "step": 4840
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0002,
+      "loss": 0.5673,
+      "step": 4850
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0002,
+      "loss": 0.6362,
+      "step": 4860
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0002,
+      "loss": 0.6337,
+      "step": 4870
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5901,
+      "step": 4880
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0002,
+      "loss": 0.6435,
+      "step": 4890
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5337,
+      "step": 4900
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5591,
+      "step": 4910
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0002,
+      "loss": 0.5847,
+      "step": 4920
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0002,
+      "loss": 0.6051,
+      "step": 4930
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002,
+      "loss": 0.5884,
+      "step": 4940
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002,
+      "loss": 0.5731,
+      "step": 4950
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5878,
+      "step": 4960
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0002,
+      "loss": 0.6194,
+      "step": 4970
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5798,
+      "step": 4980
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5841,
+      "step": 4990
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0002,
+      "loss": 0.61,
+      "step": 5000
+    },
+    {
+      "epoch": 2.78,
+      "eval_loss": 0.7293540239334106,
+      "eval_runtime": 80.9574,
+      "eval_samples_per_second": 12.352,
+      "eval_steps_per_second": 6.176,
+      "step": 5000
+    },
+    {
+      "epoch": 2.78,
+      "mmlu_eval_accuracy": 0.48996341553722794,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
+      "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_high_school_chemistry": 0.36363636363636365,
+      "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": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
+      "mmlu_eval_accuracy_high_school_statistics": 0.2608695652173913,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "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": 0.2727272727272727,
+      "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": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.28,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_professional_psychology": 0.42028985507246375,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "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": 0.6842105263157895,
+      "mmlu_loss": 1.7054601166173622,
+      "step": 5000
     }
   ],
   "max_steps": 10000,
   "num_train_epochs": 6,
-  "total_flos": 1.3617642645410611e+17,
+  "total_flos": 2.429658936706007e+17,
   "trial_name": null,
   "trial_params": null
 }