Training in progress, step 5200

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e70a0d4f464ceef8b471d0b8aadac4957200ea829c64f13b905ec09c10b0731
+oid sha256:9a7a576d8abac9c180d7d5591d9667a3c9e3eaee3e91b5175ef5df8d81f6d041
 size 319977229

{checkpoint-3000 → checkpoint-5000/adapter_model/adapter_model}/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df14a9ecce01e107eef3fff8b3daa58e6e8b0cc548cbd13f4ed4b3ec0fe05529
+oid sha256:6e70a0d4f464ceef8b471d0b8aadac4957200ea829c64f13b905ec09c10b0731
 size 319977229

checkpoint-5200/README.md

@@ -0,0 +1,20 @@
+---
+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-5200/adapter_config.json

@@ -0,0 +1,26 @@
+{
+  "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": [
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "v_proj",
+    "down_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM"
+}

checkpoint-5200/adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a7a576d8abac9c180d7d5591d9667a3c9e3eaee3e91b5175ef5df8d81f6d041
+size 319977229

{checkpoint-3000 → checkpoint-5200}/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ce2cc8fbfafdead8bf7fe167df2c1d09bcbd881c94df58a25e31a4d9348fd30
+oid sha256:02a9727e6d1556d68c116530b44d6342f22003ce214cb7e88331b98345c1241d
 size 1279539973

{checkpoint-3000 → checkpoint-5200}/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f141eafc0049f69f954b30ef3f97e5889cdf06e48933080ac8b804130ffea6ed
+oid sha256:70848f5fbc836c49c33e641ff2366550782442654d4a2c91ef7886b8b83d45a2
 size 14511

{checkpoint-3000 → checkpoint-5200}/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb57538b0fddb988f32d5e33311bcf25efee1aa4001ec3e33ef4a2dd884d77d3
+oid sha256:bfeaf01effb5693db43ee13e167cd0d73532fb4b4a2468aeb23c02cf097773a8
 size 627

{checkpoint-3000 → checkpoint-5200}/trainer_state.json

@@ -1,8 +1,8 @@
 {
-  "best_metric": 0.8454821705818176,
-  "best_model_checkpoint": "experts/expert-11/checkpoint-2600",
-  "epoch": 1.1299435028248588,
-  "global_step": 3000,
+  "best_metric": 0.8410314917564392,
+  "best_model_checkpoint": "experts/expert-11/checkpoint-5200",
+  "epoch": 1.9585687382297552,
+  "global_step": 5200,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -2871,11 +2871,2112 @@
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
       "mmlu_loss": 1.0982705684617045,
       "step": 3000
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0002,
+      "loss": 0.8004,
+      "step": 3010
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0002,
+      "loss": 0.8208,
+      "step": 3020
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0002,
+      "loss": 0.8226,
+      "step": 3030
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0002,
+      "loss": 0.8573,
+      "step": 3040
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0002,
+      "loss": 0.7801,
+      "step": 3050
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0002,
+      "loss": 0.742,
+      "step": 3060
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0002,
+      "loss": 0.8023,
+      "step": 3070
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0002,
+      "loss": 0.8149,
+      "step": 3080
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0002,
+      "loss": 0.7915,
+      "step": 3090
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0002,
+      "loss": 0.7944,
+      "step": 3100
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0002,
+      "loss": 0.8266,
+      "step": 3110
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002,
+      "loss": 0.7756,
+      "step": 3120
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002,
+      "loss": 0.8144,
+      "step": 3130
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002,
+      "loss": 0.8295,
+      "step": 3140
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002,
+      "loss": 0.7922,
+      "step": 3150
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002,
+      "loss": 0.7796,
+      "step": 3160
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002,
+      "loss": 0.7998,
+      "step": 3170
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.0002,
+      "loss": 0.8186,
+      "step": 3180
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.0002,
+      "loss": 0.7967,
+      "step": 3190
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0002,
+      "loss": 0.834,
+      "step": 3200
+    },
+    {
+      "epoch": 1.21,
+      "eval_loss": 0.848586916923523,
+      "eval_runtime": 103.5955,
+      "eval_samples_per_second": 9.653,
+      "eval_steps_per_second": 4.826,
+      "step": 3200
+    },
+    {
+      "epoch": 1.21,
+      "mmlu_eval_accuracy": 0.49133937037124614,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "mmlu_eval_accuracy_college_biology": 0.3125,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.46875,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_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": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.3953488372093023,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "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": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.72,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.6,
+      "mmlu_eval_accuracy_professional_accounting": 0.25806451612903225,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.0464347072004028,
+      "step": 3200
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0002,
+      "loss": 0.8097,
+      "step": 3210
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0002,
+      "loss": 0.8567,
+      "step": 3220
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002,
+      "loss": 0.7778,
+      "step": 3230
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002,
+      "loss": 0.7794,
+      "step": 3240
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002,
+      "loss": 0.7899,
+      "step": 3250
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0002,
+      "loss": 0.7361,
+      "step": 3260
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0002,
+      "loss": 0.7566,
+      "step": 3270
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002,
+      "loss": 0.79,
+      "step": 3280
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002,
+      "loss": 0.7555,
+      "step": 3290
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002,
+      "loss": 0.8865,
+      "step": 3300
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0002,
+      "loss": 0.8102,
+      "step": 3310
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0002,
+      "loss": 0.7986,
+      "step": 3320
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0002,
+      "loss": 0.7855,
+      "step": 3330
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.0002,
+      "loss": 0.8005,
+      "step": 3340
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.0002,
+      "loss": 0.8243,
+      "step": 3350
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0002,
+      "loss": 0.8487,
+      "step": 3360
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0002,
+      "loss": 0.7882,
+      "step": 3370
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0002,
+      "loss": 0.8584,
+      "step": 3380
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002,
+      "loss": 0.8118,
+      "step": 3390
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002,
+      "loss": 0.7495,
+      "step": 3400
+    },
+    {
+      "epoch": 1.28,
+      "eval_loss": 0.8491889834403992,
+      "eval_runtime": 103.6412,
+      "eval_samples_per_second": 9.649,
+      "eval_steps_per_second": 4.824,
+      "step": 3400
+    },
+    {
+      "epoch": 1.28,
+      "mmlu_eval_accuracy": 0.4934772222778397,
+      "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": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.1875,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "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": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3103448275862069,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.72,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.7093023255813954,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
+      "mmlu_eval_accuracy_prehistory": 0.6285714285714286,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
+      "mmlu_eval_accuracy_professional_medicine": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_psychology": 0.4492753623188406,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.45454545454545453,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.116540457498934,
+      "step": 3400
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002,
+      "loss": 0.8036,
+      "step": 3410
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.0002,
+      "loss": 0.8412,
+      "step": 3420
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.0002,
+      "loss": 0.77,
+      "step": 3430
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0002,
+      "loss": 0.8066,
+      "step": 3440
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0002,
+      "loss": 0.757,
+      "step": 3450
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0002,
+      "loss": 0.8126,
+      "step": 3460
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0002,
+      "loss": 0.7479,
+      "step": 3470
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0002,
+      "loss": 0.8028,
+      "step": 3480
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0002,
+      "loss": 0.7664,
+      "step": 3490
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002,
+      "loss": 0.7957,
+      "step": 3500
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002,
+      "loss": 0.8181,
+      "step": 3510
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0002,
+      "loss": 0.7527,
+      "step": 3520
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0002,
+      "loss": 0.7954,
+      "step": 3530
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0002,
+      "loss": 0.7975,
+      "step": 3540
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0002,
+      "loss": 0.7973,
+      "step": 3550
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0002,
+      "loss": 0.7996,
+      "step": 3560
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0002,
+      "loss": 0.7606,
+      "step": 3570
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.0002,
+      "loss": 0.852,
+      "step": 3580
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.0002,
+      "loss": 0.8273,
+      "step": 3590
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002,
+      "loss": 0.7667,
+      "step": 3600
+    },
+    {
+      "epoch": 1.36,
+      "eval_loss": 0.8506274819374084,
+      "eval_runtime": 103.589,
+      "eval_samples_per_second": 9.654,
+      "eval_steps_per_second": 4.827,
+      "step": 3600
+    },
+    {
+      "epoch": 1.36,
+      "mmlu_eval_accuracy": 0.49897089203407463,
+      "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": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.4146341463414634,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "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": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "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": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.7209302325581395,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.6285714285714286,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
+      "mmlu_eval_accuracy_professional_psychology": 0.4492753623188406,
+      "mmlu_eval_accuracy_public_relations": 0.4166666666666667,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.0956238023450418,
+      "step": 3600
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002,
+      "loss": 0.8076,
+      "step": 3610
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002,
+      "loss": 0.8269,
+      "step": 3620
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002,
+      "loss": 0.7878,
+      "step": 3630
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002,
+      "loss": 0.8249,
+      "step": 3640
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002,
+      "loss": 0.8213,
+      "step": 3650
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002,
+      "loss": 0.786,
+      "step": 3660
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002,
+      "loss": 0.7965,
+      "step": 3670
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0002,
+      "loss": 0.7713,
+      "step": 3680
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0002,
+      "loss": 0.8147,
+      "step": 3690
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0002,
+      "loss": 0.78,
+      "step": 3700
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002,
+      "loss": 0.8121,
+      "step": 3710
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002,
+      "loss": 0.7998,
+      "step": 3720
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002,
+      "loss": 0.8097,
+      "step": 3730
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002,
+      "loss": 0.8239,
+      "step": 3740
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002,
+      "loss": 0.7818,
+      "step": 3750
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002,
+      "loss": 0.7612,
+      "step": 3760
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002,
+      "loss": 0.7535,
+      "step": 3770
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002,
+      "loss": 0.8227,
+      "step": 3780
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.0002,
+      "loss": 0.794,
+      "step": 3790
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.0002,
+      "loss": 0.872,
+      "step": 3800
+    },
+    {
+      "epoch": 1.43,
+      "eval_loss": 0.8473048806190491,
+      "eval_runtime": 103.5802,
+      "eval_samples_per_second": 9.654,
+      "eval_steps_per_second": 4.827,
+      "step": 3800
+    },
+    {
+      "epoch": 1.43,
+      "mmlu_eval_accuracy": 0.4945249469225442,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5454545454545454,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.46153846153846156,
+      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.3902439024390244,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.13793103448275862,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.8181818181818182,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.6285714285714286,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_medicine": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_psychology": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.5555555555555556,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.0358139975069398,
+      "step": 3800
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002,
+      "loss": 0.7601,
+      "step": 3810
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002,
+      "loss": 0.7511,
+      "step": 3820
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002,
+      "loss": 0.7827,
+      "step": 3830
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002,
+      "loss": 0.8717,
+      "step": 3840
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002,
+      "loss": 0.7638,
+      "step": 3850
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002,
+      "loss": 0.7666,
+      "step": 3860
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002,
+      "loss": 0.8272,
+      "step": 3870
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002,
+      "loss": 0.8035,
+      "step": 3880
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002,
+      "loss": 0.8456,
+      "step": 3890
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002,
+      "loss": 0.7881,
+      "step": 3900
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002,
+      "loss": 0.8831,
+      "step": 3910
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002,
+      "loss": 0.8084,
+      "step": 3920
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002,
+      "loss": 0.7998,
+      "step": 3930
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002,
+      "loss": 0.8079,
+      "step": 3940
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.0002,
+      "loss": 0.7977,
+      "step": 3950
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.0002,
+      "loss": 0.8197,
+      "step": 3960
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002,
+      "loss": 0.8216,
+      "step": 3970
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002,
+      "loss": 0.79,
+      "step": 3980
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002,
+      "loss": 0.8336,
+      "step": 3990
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002,
+      "loss": 0.7727,
+      "step": 4000
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.8489538431167603,
+      "eval_runtime": 103.5906,
+      "eval_samples_per_second": 9.653,
+      "eval_steps_per_second": 4.827,
+      "step": 4000
+    },
+    {
+      "epoch": 1.51,
+      "mmlu_eval_accuracy": 0.500900849590307,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.7272727272727273,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "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": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "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": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5769230769230769,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
+      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.8461538461538461,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 1.0,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "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": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.5,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.5909090909090909,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.0749414411249423,
+      "step": 4000
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002,
+      "loss": 0.8238,
+      "step": 4010
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002,
+      "loss": 0.8226,
+      "step": 4020
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002,
+      "loss": 0.8221,
+      "step": 4030
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002,
+      "loss": 0.8028,
+      "step": 4040
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002,
+      "loss": 0.7944,
+      "step": 4050
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002,
+      "loss": 0.8115,
+      "step": 4060
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002,
+      "loss": 0.8019,
+      "step": 4070
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.7925,
+      "step": 4080
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.7882,
+      "step": 4090
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002,
+      "loss": 0.8062,
+      "step": 4100
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002,
+      "loss": 0.7945,
+      "step": 4110
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002,
+      "loss": 0.7377,
+      "step": 4120
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.7929,
+      "step": 4130
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.7933,
+      "step": 4140
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002,
+      "loss": 0.8386,
+      "step": 4150
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.8249,
+      "step": 4160
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.7745,
+      "step": 4170
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.0002,
+      "loss": 0.8216,
+      "step": 4180
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002,
+      "loss": 0.8174,
+      "step": 4190
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002,
+      "loss": 0.7184,
+      "step": 4200
+    },
+    {
+      "epoch": 1.58,
+      "eval_loss": 0.8472813367843628,
+      "eval_runtime": 103.5584,
+      "eval_samples_per_second": 9.656,
+      "eval_steps_per_second": 4.828,
+      "step": 4200
+    },
+    {
+      "epoch": 1.58,
+      "mmlu_eval_accuracy": 0.49769679043281745,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.5,
+      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
+      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_formal_logic": 0.35714285714285715,
+      "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": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.1724137931034483,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.35294117647058826,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8,
+      "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": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.7272727272727273,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.5263157894736842,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
+      "mmlu_eval_accuracy_nutrition": 0.5454545454545454,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
+      "mmlu_eval_accuracy_professional_psychology": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.5555555555555556,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.45454545454545453,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.057873975864901,
+      "step": 4200
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.762,
+      "step": 4210
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.8217,
+      "step": 4220
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002,
+      "loss": 0.81,
+      "step": 4230
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.7567,
+      "step": 4240
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.8548,
+      "step": 4250
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002,
+      "loss": 0.8099,
+      "step": 4260
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.8049,
+      "step": 4270
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002,
+      "loss": 0.8519,
+      "step": 4280
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.7614,
+      "step": 4290
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.783,
+      "step": 4300
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002,
+      "loss": 0.772,
+      "step": 4310
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.7742,
+      "step": 4320
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.7578,
+      "step": 4330
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.0002,
+      "loss": 0.7872,
+      "step": 4340
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.7467,
+      "step": 4350
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002,
+      "loss": 0.7874,
+      "step": 4360
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.7895,
+      "step": 4370
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.7935,
+      "step": 4380
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002,
+      "loss": 0.7907,
+      "step": 4390
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.8044,
+      "step": 4400
+    },
+    {
+      "epoch": 1.66,
+      "eval_loss": 0.8433162569999695,
+      "eval_runtime": 103.6378,
+      "eval_samples_per_second": 9.649,
+      "eval_steps_per_second": 4.824,
+      "step": 4400
+    },
+    {
+      "epoch": 1.66,
+      "mmlu_eval_accuracy": 0.4999850802460373,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
+      "mmlu_eval_accuracy_astronomy": 0.4375,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.5454545454545454,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "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": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "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": 0.8636363636363636,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3448275862068966,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.7666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_us_history": 0.7272727272727273,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6111111111111112,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.5454545454545454,
+      "mmlu_eval_accuracy_marketing": 0.72,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
+      "mmlu_eval_accuracy_moral_disputes": 0.5263157894736842,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.6176470588235294,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_professional_medicine": 0.5161290322580645,
+      "mmlu_eval_accuracy_professional_psychology": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.0007563972014049,
+      "step": 4400
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.7724,
+      "step": 4410
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002,
+      "loss": 0.7909,
+      "step": 4420
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.7663,
+      "step": 4430
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002,
+      "loss": 0.7518,
+      "step": 4440
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.7676,
+      "step": 4450
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.7993,
+      "step": 4460
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002,
+      "loss": 0.7974,
+      "step": 4470
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.8254,
+      "step": 4480
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.784,
+      "step": 4490
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002,
+      "loss": 0.8125,
+      "step": 4500
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.7862,
+      "step": 4510
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002,
+      "loss": 0.7812,
+      "step": 4520
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.7707,
+      "step": 4530
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.8346,
+      "step": 4540
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002,
+      "loss": 0.7844,
+      "step": 4550
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002,
+      "loss": 0.7777,
+      "step": 4560
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002,
+      "loss": 0.8289,
+      "step": 4570
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.8437,
+      "step": 4580
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.7947,
+      "step": 4590
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002,
+      "loss": 0.8127,
+      "step": 4600
+    },
+    {
+      "epoch": 1.73,
+      "eval_loss": 0.8427039980888367,
+      "eval_runtime": 103.6307,
+      "eval_samples_per_second": 9.65,
+      "eval_steps_per_second": 4.825,
+      "step": 4600
+    },
+    {
+      "epoch": 1.73,
+      "mmlu_eval_accuracy": 0.4940456449423768,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_astronomy": 0.375,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "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": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.45454545454545453,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3448275862068966,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
+      "mmlu_eval_accuracy_high_school_psychology": 0.7833333333333333,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
+      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.3235294117647059,
+      "mmlu_eval_accuracy_professional_medicine": 0.4838709677419355,
+      "mmlu_eval_accuracy_professional_psychology": 0.42028985507246375,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.0355537600873657,
+      "step": 4600
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.8198,
+      "step": 4610
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.8591,
+      "step": 4620
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002,
+      "loss": 0.8531,
+      "step": 4630
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.8161,
+      "step": 4640
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002,
+      "loss": 0.7697,
+      "step": 4650
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.7589,
+      "step": 4660
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.7917,
+      "step": 4670
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 0.7926,
+      "step": 4680
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.8375,
+      "step": 4690
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.8241,
+      "step": 4700
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0002,
+      "loss": 0.7863,
+      "step": 4710
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.8374,
+      "step": 4720
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002,
+      "loss": 0.8026,
+      "step": 4730
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.8514,
+      "step": 4740
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.7902,
+      "step": 4750
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002,
+      "loss": 0.8603,
+      "step": 4760
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.7253,
+      "step": 4770
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.8327,
+      "step": 4780
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002,
+      "loss": 0.8101,
+      "step": 4790
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.8036,
+      "step": 4800
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.843153178691864,
+      "eval_runtime": 103.636,
+      "eval_samples_per_second": 9.649,
+      "eval_steps_per_second": 4.825,
+      "step": 4800
+    },
+    {
+      "epoch": 1.81,
+      "mmlu_eval_accuracy": 0.5086466339912666,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.6363636363636364,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "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": 0.2727272727272727,
+      "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": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5769230769230769,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
+      "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
+      "mmlu_eval_accuracy_international_law": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.76,
+      "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": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.6176470588235294,
+      "mmlu_eval_accuracy_prehistory": 0.6,
+      "mmlu_eval_accuracy_professional_accounting": 0.2903225806451613,
+      "mmlu_eval_accuracy_professional_law": 0.34705882352941175,
+      "mmlu_eval_accuracy_professional_medicine": 0.4838709677419355,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.6111111111111112,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.0574546023892049,
+      "step": 4800
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002,
+      "loss": 0.803,
+      "step": 4810
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.7996,
+      "step": 4820
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.7841,
+      "step": 4830
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002,
+      "loss": 0.871,
+      "step": 4840
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.7837,
+      "step": 4850
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.802,
+      "step": 4860
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002,
+      "loss": 0.8178,
+      "step": 4870
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.771,
+      "step": 4880
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002,
+      "loss": 0.7621,
+      "step": 4890
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.8036,
+      "step": 4900
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.7915,
+      "step": 4910
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002,
+      "loss": 0.7525,
+      "step": 4920
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.8213,
+      "step": 4930
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.8155,
+      "step": 4940
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0002,
+      "loss": 0.7462,
+      "step": 4950
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002,
+      "loss": 0.8139,
+      "step": 4960
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002,
+      "loss": 0.8439,
+      "step": 4970
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.7555,
+      "step": 4980
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.829,
+      "step": 4990
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002,
+      "loss": 0.8246,
+      "step": 5000
+    },
+    {
+      "epoch": 1.88,
+      "eval_loss": 0.8419777750968933,
+      "eval_runtime": 103.5463,
+      "eval_samples_per_second": 9.658,
+      "eval_steps_per_second": 4.829,
+      "step": 5000
+    },
+    {
+      "epoch": 1.88,
+      "mmlu_eval_accuracy": 0.5123524674376934,
+      "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "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": 0.25,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
+      "mmlu_eval_accuracy_formal_logic": 0.2857142857142857,
+      "mmlu_eval_accuracy_global_facts": 0.6,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+      "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3448275862068966,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
+      "mmlu_eval_accuracy_management": 0.8181818181818182,
+      "mmlu_eval_accuracy_marketing": 0.72,
+      "mmlu_eval_accuracy_medical_genetics": 0.8181818181818182,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.6176470588235294,
+      "mmlu_eval_accuracy_prehistory": 0.5714285714285714,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.35294117647058826,
+      "mmlu_eval_accuracy_professional_medicine": 0.45161290322580644,
+      "mmlu_eval_accuracy_professional_psychology": 0.43478260869565216,
+      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
+      "mmlu_eval_accuracy_security_studies": 0.5555555555555556,
+      "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.1149214952406,
+      "step": 5000
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.7821,
+      "step": 5010
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.8108,
+      "step": 5020
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002,
+      "loss": 0.7921,
+      "step": 5030
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.7999,
+      "step": 5040
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002,
+      "loss": 0.8019,
+      "step": 5050
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.7978,
+      "step": 5060
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.7798,
+      "step": 5070
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002,
+      "loss": 0.8006,
+      "step": 5080
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.8297,
+      "step": 5090
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.7629,
+      "step": 5100
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002,
+      "loss": 0.7935,
+      "step": 5110
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.8167,
+      "step": 5120
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002,
+      "loss": 0.7555,
+      "step": 5130
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.7798,
+      "step": 5140
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.7948,
+      "step": 5150
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002,
+      "loss": 0.8189,
+      "step": 5160
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.8086,
+      "step": 5170
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.7941,
+      "step": 5180
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002,
+      "loss": 0.7684,
+      "step": 5190
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002,
+      "loss": 0.789,
+      "step": 5200
+    },
+    {
+      "epoch": 1.96,
+      "eval_loss": 0.8410314917564392,
+      "eval_runtime": 104.2734,
+      "eval_samples_per_second": 9.59,
+      "eval_steps_per_second": 4.795,
+      "step": 5200
+    },
+    {
+      "epoch": 1.96,
+      "mmlu_eval_accuracy": 0.5087413100683031,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "mmlu_eval_accuracy_astronomy": 0.5,
+      "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.41379310344827586,
+      "mmlu_eval_accuracy_college_biology": 0.375,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
+      "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_mathematics": 0.18181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
+      "mmlu_eval_accuracy_conceptual_physics": 0.3076923076923077,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
+      "mmlu_eval_accuracy_electrical_engineering": 0.1875,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "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": 0.5555555555555556,
+      "mmlu_eval_accuracy_high_school_geography": 0.8181818181818182,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.3103448275862069,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
+      "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8166666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
+      "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": 0.9230769230769231,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
+      "mmlu_eval_accuracy_management": 0.6363636363636364,
+      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_medical_genetics": 1.0,
+      "mmlu_eval_accuracy_miscellaneous": 0.6395348837209303,
+      "mmlu_eval_accuracy_moral_disputes": 0.5263157894736842,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
+      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
+      "mmlu_eval_accuracy_professional_law": 0.3588235294117647,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.42028985507246375,
+      "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
+      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
+      "mmlu_eval_accuracy_virology": 0.5,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.1855050965015321,
+      "step": 5200
     }
   ],
   "max_steps": 10000,
   "num_train_epochs": 4,
-  "total_flos": 8.69041954700329e+17,
+  "total_flos": 1.5049528999840973e+18,
   "trial_name": null,
   "trial_params": null
 }