Training in progress, step 5600

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e959ec076b70327ea5ec0ffcbd8223613a1afd80888e03cd9d40e3a2595508b
+oid sha256:9934fea07d489e8ae91846e04370c9655e86a84523347c29e93ea7eeeaa09b24
 size 319977229

checkpoint-3400/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-3400/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": [
-    "q_proj",
-    "up_proj",
-    "gate_proj",
-    "k_proj",
-    "o_proj",
-    "v_proj",
-    "down_proj"
-  ],
-  "task_type": "CAUSAL_LM"
-}

checkpoint-3400/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:eeb65691ef5962d317c8efec7d0b79cb87640a7b978650a743b1e2bdbaa0a2d5
-size 319977229

checkpoint-5200/adapter_model/adapter_model/README.md

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

checkpoint-5200/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d378ad66e49132f656022e64847dfa1dbe6ad496d987305e79866a75f22d17c8
+oid sha256:9e959ec076b70327ea5ec0ffcbd8223613a1afd80888e03cd9d40e3a2595508b
 size 319977229

{checkpoint-3400 → checkpoint-5600}/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eeb65691ef5962d317c8efec7d0b79cb87640a7b978650a743b1e2bdbaa0a2d5
+oid sha256:9934fea07d489e8ae91846e04370c9655e86a84523347c29e93ea7eeeaa09b24
 size 319977229

{checkpoint-3400 → checkpoint-5600}/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48cb84254bb05874aafa2a0ac083d5240fdfc993178f1435f74ced0fc0c086db
+oid sha256:6caed951afccace98899cf8db5b76c9160cec7c63b0f282a8f907060feab1e6d
 size 1279539973

{checkpoint-3400 → checkpoint-5600}/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e8e54ce8d8790b73966644ab99d7d11ce1baf113a5f5bebb5f8aca16d87f677
+oid sha256:0c47faa9d8708f79f5f68f108f5324f55dab5ccb7d79f10a0ec7e296d7a37c27
 size 14511

{checkpoint-3400 → checkpoint-5600}/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb3309524185001782da002e1efb7406e18966db2aee795c24a16a9ee18dd8d0
+oid sha256:ba9e18c7c96a1432f6ee82e99802fc56a2dd7ef9d1f20dda433c9b754ac0514c
 size 627

{checkpoint-3400 → checkpoint-5600}/trainer_state.json

@@ -1,8 +1,8 @@
 {
-  "best_metric": 0.5195473432540894,
-  "best_model_checkpoint": "experts/expert-12/checkpoint-3400",
-  "epoch": 0.6064119142105497,
-  "global_step": 3400,
+  "best_metric": 0.5056349635124207,
+  "best_model_checkpoint": "experts/expert-12/checkpoint-5600",
+  "epoch": 0.9987960939938467,
+  "global_step": 5600,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -3253,11 +3253,2112 @@
       "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
       "mmlu_loss": 1.336662752899115,
       "step": 3400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002,
+      "loss": 0.5168,
+      "step": 3410
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002,
+      "loss": 0.5364,
+      "step": 3420
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002,
+      "loss": 0.6188,
+      "step": 3430
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002,
+      "loss": 0.5455,
+      "step": 3440
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5688,
+      "step": 3450
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5037,
+      "step": 3460
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5701,
+      "step": 3470
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5184,
+      "step": 3480
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5695,
+      "step": 3490
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002,
+      "loss": 0.5325,
+      "step": 3500
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5066,
+      "step": 3510
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5482,
+      "step": 3520
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5311,
+      "step": 3530
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5397,
+      "step": 3540
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.5678,
+      "step": 3550
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002,
+      "loss": 0.513,
+      "step": 3560
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5331,
+      "step": 3570
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5447,
+      "step": 3580
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5683,
+      "step": 3590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5266,
+      "step": 3600
+    },
+    {
+      "epoch": 0.64,
+      "eval_loss": 0.5180239081382751,
+      "eval_runtime": 102.2373,
+      "eval_samples_per_second": 9.781,
+      "eval_steps_per_second": 4.891,
+      "step": 3600
+    },
+    {
+      "epoch": 0.64,
+      "mmlu_eval_accuracy": 0.496455891933492,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "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": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "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": 0.6956521739130435,
+      "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": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.3684210526315789,
+      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "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": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "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": 0.7368421052631579,
+      "mmlu_loss": 1.3701581433609944,
+      "step": 3600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002,
+      "loss": 0.5192,
+      "step": 3610
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.531,
+      "step": 3620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.5074,
+      "step": 3630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.539,
+      "step": 3640
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.5645,
+      "step": 3650
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.4887,
+      "step": 3660
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002,
+      "loss": 0.5888,
+      "step": 3670
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5287,
+      "step": 3680
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5329,
+      "step": 3690
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5768,
+      "step": 3700
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5311,
+      "step": 3710
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002,
+      "loss": 0.5618,
+      "step": 3720
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5543,
+      "step": 3730
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5959,
+      "step": 3740
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.4902,
+      "step": 3750
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5645,
+      "step": 3760
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5583,
+      "step": 3770
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002,
+      "loss": 0.5349,
+      "step": 3780
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5001,
+      "step": 3790
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5429,
+      "step": 3800
+    },
+    {
+      "epoch": 0.68,
+      "eval_loss": 0.5150953531265259,
+      "eval_runtime": 102.3224,
+      "eval_samples_per_second": 9.773,
+      "eval_steps_per_second": 4.887,
+      "step": 3800
+    },
+    {
+      "epoch": 0.68,
+      "mmlu_eval_accuracy": 0.5026698194241785,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+      "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": 0.4090909090909091,
+      "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "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": 0.375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.34615384615384615,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.4782608695652174,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.782608695652174,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_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": 1.0,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.3684210526315789,
+      "mmlu_eval_accuracy_moral_scenarios": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "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": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.2356752671165192,
+      "step": 3800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5206,
+      "step": 3810
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5707,
+      "step": 3820
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5143,
+      "step": 3830
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002,
+      "loss": 0.5968,
+      "step": 3840
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002,
+      "loss": 0.4984,
+      "step": 3850
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002,
+      "loss": 0.5292,
+      "step": 3860
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002,
+      "loss": 0.5169,
+      "step": 3870
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002,
+      "loss": 0.5284,
+      "step": 3880
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002,
+      "loss": 0.4888,
+      "step": 3890
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.4798,
+      "step": 3900
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.526,
+      "step": 3910
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.5376,
+      "step": 3920
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.5237,
+      "step": 3930
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.5684,
+      "step": 3940
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002,
+      "loss": 0.5651,
+      "step": 3950
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002,
+      "loss": 0.5232,
+      "step": 3960
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002,
+      "loss": 0.5465,
+      "step": 3970
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002,
+      "loss": 0.5362,
+      "step": 3980
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002,
+      "loss": 0.5124,
+      "step": 3990
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002,
+      "loss": 0.5179,
+      "step": 4000
+    },
+    {
+      "epoch": 0.71,
+      "eval_loss": 0.5137460231781006,
+      "eval_runtime": 102.3369,
+      "eval_samples_per_second": 9.772,
+      "eval_steps_per_second": 4.886,
+      "step": 4000
+    },
+    {
+      "epoch": 0.71,
+      "mmlu_eval_accuracy": 0.494439834261674,
+      "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": 0.5862068965517241,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.35714285714285715,
+      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "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": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "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": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.2,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.32444475490026,
+      "step": 4000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.4896,
+      "step": 4010
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5238,
+      "step": 4020
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5296,
+      "step": 4030
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5339,
+      "step": 4040
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5237,
+      "step": 4050
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002,
+      "loss": 0.5221,
+      "step": 4060
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5329,
+      "step": 4070
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.4942,
+      "step": 4080
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.4832,
+      "step": 4090
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5202,
+      "step": 4100
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5808,
+      "step": 4110
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002,
+      "loss": 0.5421,
+      "step": 4120
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002,
+      "loss": 0.5135,
+      "step": 4130
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002,
+      "loss": 0.556,
+      "step": 4140
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002,
+      "loss": 0.5764,
+      "step": 4150
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002,
+      "loss": 0.5991,
+      "step": 4160
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002,
+      "loss": 0.5573,
+      "step": 4170
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.545,
+      "step": 4180
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.5707,
+      "step": 4190
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.5406,
+      "step": 4200
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.5133056640625,
+      "eval_runtime": 102.4142,
+      "eval_samples_per_second": 9.764,
+      "eval_steps_per_second": 4.882,
+      "step": 4200
+    },
+    {
+      "epoch": 0.75,
+      "mmlu_eval_accuracy": 0.4939276831125043,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "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": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.4,
+      "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": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "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": 0.3333333333333333,
+      "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": 0.09090909090909091,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.3083451582464163,
+      "step": 4200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.5343,
+      "step": 4210
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.4992,
+      "step": 4220
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002,
+      "loss": 0.4969,
+      "step": 4230
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002,
+      "loss": 0.6008,
+      "step": 4240
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5667,
+      "step": 4250
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5312,
+      "step": 4260
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5436,
+      "step": 4270
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002,
+      "loss": 0.5367,
+      "step": 4280
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5298,
+      "step": 4290
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5148,
+      "step": 4300
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5503,
+      "step": 4310
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5568,
+      "step": 4320
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5378,
+      "step": 4330
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002,
+      "loss": 0.5268,
+      "step": 4340
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5356,
+      "step": 4350
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5296,
+      "step": 4360
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5251,
+      "step": 4370
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5216,
+      "step": 4380
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5477,
+      "step": 4390
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002,
+      "loss": 0.5498,
+      "step": 4400
+    },
+    {
+      "epoch": 0.78,
+      "eval_loss": 0.5130004286766052,
+      "eval_runtime": 102.4487,
+      "eval_samples_per_second": 9.761,
+      "eval_steps_per_second": 4.88,
+      "step": 4400
+    },
+    {
+      "epoch": 0.78,
+      "mmlu_eval_accuracy": 0.4929320314097983,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "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": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "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": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6111111111111112,
+      "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": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.2,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.273639883291628,
+      "step": 4400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002,
+      "loss": 0.5042,
+      "step": 4410
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002,
+      "loss": 0.5773,
+      "step": 4420
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002,
+      "loss": 0.519,
+      "step": 4430
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002,
+      "loss": 0.5431,
+      "step": 4440
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002,
+      "loss": 0.5039,
+      "step": 4450
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5874,
+      "step": 4460
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5504,
+      "step": 4470
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5289,
+      "step": 4480
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5295,
+      "step": 4490
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.478,
+      "step": 4500
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002,
+      "loss": 0.5006,
+      "step": 4510
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002,
+      "loss": 0.5207,
+      "step": 4520
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002,
+      "loss": 0.4869,
+      "step": 4530
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002,
+      "loss": 0.5574,
+      "step": 4540
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002,
+      "loss": 0.5191,
+      "step": 4550
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002,
+      "loss": 0.5378,
+      "step": 4560
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.5425,
+      "step": 4570
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.556,
+      "step": 4580
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.5145,
+      "step": 4590
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.5405,
+      "step": 4600
+    },
+    {
+      "epoch": 0.82,
+      "eval_loss": 0.5101557374000549,
+      "eval_runtime": 102.4142,
+      "eval_samples_per_second": 9.764,
+      "eval_steps_per_second": 4.882,
+      "step": 4600
+    },
+    {
+      "epoch": 0.82,
+      "mmlu_eval_accuracy": 0.4853678817693758,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "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": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.3181818181818182,
+      "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.5,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_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": 0.6666666666666666,
+      "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": 0.8833333333333333,
+      "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": 0.6521739130434783,
+      "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": 0.09090909090909091,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
+      "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": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.4017060332525493,
+      "step": 4600
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.5307,
+      "step": 4610
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002,
+      "loss": 0.5598,
+      "step": 4620
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5214,
+      "step": 4630
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5466,
+      "step": 4640
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5417,
+      "step": 4650
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5349,
+      "step": 4660
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5859,
+      "step": 4670
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002,
+      "loss": 0.5071,
+      "step": 4680
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002,
+      "loss": 0.5443,
+      "step": 4690
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002,
+      "loss": 0.574,
+      "step": 4700
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002,
+      "loss": 0.5311,
+      "step": 4710
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002,
+      "loss": 0.5195,
+      "step": 4720
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002,
+      "loss": 0.5464,
+      "step": 4730
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5169,
+      "step": 4740
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5427,
+      "step": 4750
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5578,
+      "step": 4760
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5147,
+      "step": 4770
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5675,
+      "step": 4780
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002,
+      "loss": 0.5303,
+      "step": 4790
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002,
+      "loss": 0.5736,
+      "step": 4800
+    },
+    {
+      "epoch": 0.86,
+      "eval_loss": 0.5110376477241516,
+      "eval_runtime": 102.4229,
+      "eval_samples_per_second": 9.763,
+      "eval_steps_per_second": 4.882,
+      "step": 4800
+    },
+    {
+      "epoch": 0.86,
+      "mmlu_eval_accuracy": 0.5007327736795305,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5862068965517241,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "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": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "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": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "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": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_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": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.27,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.192659597950878,
+      "step": 4800
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002,
+      "loss": 0.5271,
+      "step": 4810
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002,
+      "loss": 0.5772,
+      "step": 4820
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002,
+      "loss": 0.5121,
+      "step": 4830
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002,
+      "loss": 0.5232,
+      "step": 4840
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.508,
+      "step": 4850
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.5485,
+      "step": 4860
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.544,
+      "step": 4870
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.5319,
+      "step": 4880
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.5319,
+      "step": 4890
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002,
+      "loss": 0.5125,
+      "step": 4900
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.511,
+      "step": 4910
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.5585,
+      "step": 4920
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.5333,
+      "step": 4930
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.4743,
+      "step": 4940
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.5057,
+      "step": 4950
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002,
+      "loss": 0.4786,
+      "step": 4960
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002,
+      "loss": 0.5109,
+      "step": 4970
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002,
+      "loss": 0.5441,
+      "step": 4980
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002,
+      "loss": 0.5542,
+      "step": 4990
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002,
+      "loss": 0.5178,
+      "step": 5000
+    },
+    {
+      "epoch": 0.89,
+      "eval_loss": 0.5084454417228699,
+      "eval_runtime": 102.455,
+      "eval_samples_per_second": 9.76,
+      "eval_steps_per_second": 4.88,
+      "step": 5000
+    },
+    {
+      "epoch": 0.89,
+      "mmlu_eval_accuracy": 0.4934526076511491,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "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": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "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": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "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": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
+      "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": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.782608695652174,
+      "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": 0.76,
+      "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": 0.22,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.334557003825514,
+      "step": 5000
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002,
+      "loss": 0.5603,
+      "step": 5010
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.53,
+      "step": 5020
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.522,
+      "step": 5030
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.5504,
+      "step": 5040
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.565,
+      "step": 5050
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.5088,
+      "step": 5060
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002,
+      "loss": 0.4826,
+      "step": 5070
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.5275,
+      "step": 5080
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.5166,
+      "step": 5090
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.5449,
+      "step": 5100
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.5048,
+      "step": 5110
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.4882,
+      "step": 5120
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002,
+      "loss": 0.5304,
+      "step": 5130
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002,
+      "loss": 0.489,
+      "step": 5140
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002,
+      "loss": 0.5581,
+      "step": 5150
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002,
+      "loss": 0.5084,
+      "step": 5160
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002,
+      "loss": 0.5148,
+      "step": 5170
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002,
+      "loss": 0.5069,
+      "step": 5180
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.486,
+      "step": 5190
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.4827,
+      "step": 5200
+    },
+    {
+      "epoch": 0.93,
+      "eval_loss": 0.5077379941940308,
+      "eval_runtime": 102.3699,
+      "eval_samples_per_second": 9.768,
+      "eval_steps_per_second": 4.884,
+      "step": 5200
+    },
+    {
+      "epoch": 0.93,
+      "mmlu_eval_accuracy": 0.49601352955039174,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5625,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "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": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "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": 0.07142857142857142,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "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": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "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": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.2727272727272727,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "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": 0.23,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "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": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.217960480688135,
+      "step": 5200
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.5066,
+      "step": 5210
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.5802,
+      "step": 5220
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.5359,
+      "step": 5230
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002,
+      "loss": 0.5458,
+      "step": 5240
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5206,
+      "step": 5250
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5386,
+      "step": 5260
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5588,
+      "step": 5270
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5311,
+      "step": 5280
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002,
+      "loss": 0.5346,
+      "step": 5290
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5085,
+      "step": 5300
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5517,
+      "step": 5310
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5421,
+      "step": 5320
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5511,
+      "step": 5330
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5875,
+      "step": 5340
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002,
+      "loss": 0.5184,
+      "step": 5350
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.5095,
+      "step": 5360
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.5811,
+      "step": 5370
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.5363,
+      "step": 5380
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.572,
+      "step": 5390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.5029,
+      "step": 5400
+    },
+    {
+      "epoch": 0.96,
+      "eval_loss": 0.5077568888664246,
+      "eval_runtime": 102.4099,
+      "eval_samples_per_second": 9.765,
+      "eval_steps_per_second": 4.882,
+      "step": 5400
+    },
+    {
+      "epoch": 0.96,
+      "mmlu_eval_accuracy": 0.49314919517228223,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5517241379310345,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
+      "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": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "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": 0.6666666666666666,
+      "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": 0.6511627906976745,
+      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.7272727272727273,
+      "mmlu_eval_accuracy_philosophy": 0.5882352941176471,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "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": 0.5,
+      "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": 0.4444444444444444,
+      "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+      "mmlu_loss": 1.264488997292892,
+      "step": 5400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002,
+      "loss": 0.5247,
+      "step": 5410
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002,
+      "loss": 0.4931,
+      "step": 5420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002,
+      "loss": 0.4669,
+      "step": 5430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002,
+      "loss": 0.5494,
+      "step": 5440
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002,
+      "loss": 0.503,
+      "step": 5450
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002,
+      "loss": 0.5399,
+      "step": 5460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.5811,
+      "step": 5470
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.47,
+      "step": 5480
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.5358,
+      "step": 5490
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.5278,
+      "step": 5500
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.45,
+      "step": 5510
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002,
+      "loss": 0.5244,
+      "step": 5520
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002,
+      "loss": 0.5073,
+      "step": 5530
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002,
+      "loss": 0.5071,
+      "step": 5540
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002,
+      "loss": 0.4513,
+      "step": 5550
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002,
+      "loss": 0.5269,
+      "step": 5560
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002,
+      "loss": 0.5342,
+      "step": 5570
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0002,
+      "loss": 0.5182,
+      "step": 5580
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0002,
+      "loss": 0.4731,
+      "step": 5590
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0002,
+      "loss": 0.5468,
+      "step": 5600
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.5056349635124207,
+      "eval_runtime": 102.3725,
+      "eval_samples_per_second": 9.768,
+      "eval_steps_per_second": 4.884,
+      "step": 5600
+    },
+    {
+      "epoch": 1.0,
+      "mmlu_eval_accuracy": 0.5000700682628523,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "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": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5454545454545454,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.375,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "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": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.782608695652174,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.2,
+      "mmlu_eval_accuracy_nutrition": 0.7575757575757576,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "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": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.326268656407884,
+      "step": 5600
     }
   ],
   "max_steps": 10000,
   "num_train_epochs": 2,
-  "total_flos": 9.363160645814845e+17,
+  "total_flos": 1.5463684000029082e+18,
   "trial_name": null,
   "trial_params": null
 }