Training in progress, step 600

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e3cbdc2396034eea056c94b711b2e61ed6b5b3342d49c60e9a941edfb9e87ac
+oid sha256:7582188887c3e9877f33180a1924745623e9f26b473bb67df1455ab5a4faa14c
 size 319977229

checkpoint-400/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-400/adapter_model/adapter_model/adapter_config.json

@@ -14,13 +14,13 @@
   "r": 64,
   "revision": null,
   "target_modules": [
+    "down_proj",
     "q_proj",
-    "o_proj",
+    "up_proj",
     "v_proj",
-    "gate_proj",
-    "down_proj",
+    "o_proj",
     "k_proj",
-    "up_proj"
+    "gate_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

checkpoint-400/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79d05eb1b295dd237c9b4a0bcd5a6a7d1fce967dae6832bbdf1ff1f0001593e8
+oid sha256:1e3cbdc2396034eea056c94b711b2e61ed6b5b3342d49c60e9a941edfb9e87ac
 size 319977229

checkpoint-600/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-600/adapter_config.json

@@ -14,13 +14,13 @@
   "r": 64,
   "revision": null,
   "target_modules": [
+    "down_proj",
     "q_proj",
-    "o_proj",
+    "up_proj",
     "v_proj",
-    "gate_proj",
-    "down_proj",
+    "o_proj",
     "k_proj",
-    "up_proj"
+    "gate_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

checkpoint-600/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc3dcf962040cbcdc8468ca5cb240852bfc3237b4941586f7ca448e64b99ee19
+oid sha256:7582188887c3e9877f33180a1924745623e9f26b473bb67df1455ab5a4faa14c
 size 319977229

checkpoint-600/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:202cb375493d170f9b38a8fa91478911ab96c12276b6f81c5106f8ebe3352dd3
+oid sha256:945f2ee32f86159988dc6fdbdf62f7d6317536d6d62f36c3fee609787700576d
 size 1279539973

checkpoint-600/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7aa80f7820b3fb08be0c80daa38f913ed89b3a9be1f89e682c36ba549310548
+oid sha256:325ecf8978046b5d65c6fce52816144b03f009b3bd5508cd55f1c27df4c7f841
 size 14511

checkpoint-600/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 0.7934040427207947,
+  "best_metric": 0.806236207485199,
   "best_model_checkpoint": "experts/expert-21/checkpoint-600",
-  "epoch": 0.23952095808383234,
+  "epoch": 0.11976047904191617,
   "global_step": 600,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
@@ -10,161 +10,161 @@
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.9029,
+      "loss": 0.9541,
       "step": 10
     },
     {
-      "epoch": 0.01,
+      "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.8429,
+      "loss": 0.8555,
       "step": 20
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8963,
+      "loss": 0.8687,
       "step": 30
     },
     {
-      "epoch": 0.02,
+      "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.9288,
+      "loss": 0.8501,
       "step": 40
     },
     {
-      "epoch": 0.02,
+      "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8166,
+      "loss": 0.863,
       "step": 50
     },
     {
-      "epoch": 0.02,
+      "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8588,
+      "loss": 0.9269,
       "step": 60
     },
     {
-      "epoch": 0.03,
+      "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8676,
+      "loss": 0.8752,
       "step": 70
     },
     {
-      "epoch": 0.03,
+      "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8118,
+      "loss": 0.9716,
       "step": 80
     },
     {
-      "epoch": 0.04,
+      "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8479,
+      "loss": 0.859,
       "step": 90
     },
     {
-      "epoch": 0.04,
+      "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8755,
+      "loss": 0.8005,
       "step": 100
     },
     {
-      "epoch": 0.04,
+      "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8899,
+      "loss": 0.9085,
       "step": 110
     },
     {
-      "epoch": 0.05,
+      "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.831,
+      "loss": 0.8209,
       "step": 120
     },
     {
-      "epoch": 0.05,
+      "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.7856,
+      "loss": 0.9187,
       "step": 130
     },
     {
-      "epoch": 0.06,
+      "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.7811,
+      "loss": 0.8224,
       "step": 140
     },
     {
-      "epoch": 0.06,
+      "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.786,
+      "loss": 0.8602,
       "step": 150
     },
     {
-      "epoch": 0.06,
+      "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8194,
+      "loss": 0.7885,
       "step": 160
     },
     {
-      "epoch": 0.07,
+      "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8858,
+      "loss": 0.821,
       "step": 170
     },
     {
-      "epoch": 0.07,
+      "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8421,
+      "loss": 0.8676,
       "step": 180
     },
     {
-      "epoch": 0.08,
+      "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8294,
+      "loss": 0.9202,
       "step": 190
     },
     {
-      "epoch": 0.08,
+      "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.877,
+      "loss": 0.8188,
       "step": 200
     },
     {
-      "epoch": 0.08,
-      "eval_loss": 0.8124681711196899,
-      "eval_runtime": 137.9176,
-      "eval_samples_per_second": 7.251,
-      "eval_steps_per_second": 3.625,
+      "epoch": 0.04,
+      "eval_loss": 0.823621928691864,
+      "eval_runtime": 189.8157,
+      "eval_samples_per_second": 5.268,
+      "eval_steps_per_second": 2.634,
       "step": 200
     },
     {
-      "epoch": 0.08,
-      "mmlu_eval_accuracy": 0.4873261919391153,
+      "epoch": 0.04,
+      "mmlu_eval_accuracy": 0.486383698866548,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
-      "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+      "mmlu_eval_accuracy_anatomy": 0.5714285714285714,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "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": 0.2727272727272727,
-      "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
-      "mmlu_eval_accuracy_college_physics": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
       "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
       "mmlu_eval_accuracy_econometrics": 0.25,
-      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
       "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
-      "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
       "mmlu_eval_accuracy_global_facts": 0.4,
-      "mmlu_eval_accuracy_high_school_biology": 0.375,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.45454545454545453,
+      "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": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
       "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.34615384615384615,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
       "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
       "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
@@ -173,195 +173,195 @@
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.9230769230769231,
-      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_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": 0.84,
+      "mmlu_eval_accuracy_marketing": 0.88,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
-      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.22,
-      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
-      "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
-      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
-      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
       "mmlu_eval_accuracy_professional_law": 0.32941176470588235,
-      "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
-      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
+      "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
+      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
-      "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
+      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
       "mmlu_eval_accuracy_sociology": 0.6363636363636364,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
-      "mmlu_loss": 1.6638376168887234,
+      "mmlu_loss": 1.6085881815722343,
       "step": 200
     },
     {
-      "epoch": 0.08,
+      "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8558,
+      "loss": 0.859,
       "step": 210
     },
     {
-      "epoch": 0.09,
+      "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8541,
+      "loss": 0.9367,
       "step": 220
     },
     {
-      "epoch": 0.09,
+      "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.7311,
+      "loss": 0.8414,
       "step": 230
     },
     {
-      "epoch": 0.1,
+      "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8155,
+      "loss": 0.8446,
       "step": 240
     },
     {
-      "epoch": 0.1,
+      "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8301,
+      "loss": 0.8391,
       "step": 250
     },
     {
-      "epoch": 0.1,
+      "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8522,
+      "loss": 0.7457,
       "step": 260
     },
     {
-      "epoch": 0.11,
+      "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.7884,
+      "loss": 0.7707,
       "step": 270
     },
     {
-      "epoch": 0.11,
+      "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8282,
+      "loss": 0.7892,
       "step": 280
     },
     {
-      "epoch": 0.12,
+      "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.849,
+      "loss": 0.8066,
       "step": 290
     },
     {
-      "epoch": 0.12,
+      "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.7855,
+      "loss": 0.8116,
       "step": 300
     },
     {
-      "epoch": 0.12,
+      "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.781,
+      "loss": 0.8572,
       "step": 310
     },
     {
-      "epoch": 0.13,
+      "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.7483,
+      "loss": 0.7915,
       "step": 320
     },
     {
-      "epoch": 0.13,
+      "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8754,
+      "loss": 0.9248,
       "step": 330
     },
     {
-      "epoch": 0.14,
+      "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8535,
+      "loss": 0.8565,
       "step": 340
     },
     {
-      "epoch": 0.14,
+      "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8352,
+      "loss": 0.8508,
       "step": 350
     },
     {
-      "epoch": 0.14,
+      "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.7893,
+      "loss": 0.8534,
       "step": 360
     },
     {
-      "epoch": 0.15,
+      "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8107,
+      "loss": 0.8604,
       "step": 370
     },
     {
-      "epoch": 0.15,
+      "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.751,
+      "loss": 0.803,
       "step": 380
     },
     {
-      "epoch": 0.16,
+      "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8323,
+      "loss": 0.9193,
       "step": 390
     },
     {
-      "epoch": 0.16,
+      "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7828,
+      "loss": 0.8465,
       "step": 400
     },
     {
-      "epoch": 0.16,
-      "eval_loss": 0.8003763556480408,
-      "eval_runtime": 137.7438,
-      "eval_samples_per_second": 7.26,
-      "eval_steps_per_second": 3.63,
+      "epoch": 0.08,
+      "eval_loss": 0.8140458464622498,
+      "eval_runtime": 187.0783,
+      "eval_samples_per_second": 5.345,
+      "eval_steps_per_second": 2.673,
       "step": 400
     },
     {
-      "epoch": 0.16,
-      "mmlu_eval_accuracy": 0.4961129097533734,
-      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "epoch": 0.08,
+      "mmlu_eval_accuracy": 0.49107726483887787,
+      "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": 0.5172413793103449,
       "mmlu_eval_accuracy_college_biology": 0.4375,
-      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "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": 0.36363636363636365,
-      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
       "mmlu_eval_accuracy_econometrics": 0.25,
-      "mmlu_eval_accuracy_electrical_engineering": 0.25,
+      "mmlu_eval_accuracy_electrical_engineering": 0.3125,
       "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
       "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
       "mmlu_eval_accuracy_global_facts": 0.4,
-      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "mmlu_eval_accuracy_high_school_biology": 0.375,
       "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.5,
-      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.34615384615384615,
+      "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
       "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
-      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
-      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
-      "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
+      "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": 0.9230769230769231,
       "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
@@ -370,189 +370,189 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.8,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
-      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
-      "mmlu_eval_accuracy_moral_scenarios": 0.2,
-      "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
-      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
-      "mmlu_eval_accuracy_prehistory": 0.42857142857142855,
-      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
-      "mmlu_eval_accuracy_professional_law": 0.3,
-      "mmlu_eval_accuracy_professional_medicine": 0.6451612903225806,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
+      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "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": 0.5806451612903226,
       "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
-      "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
-      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
-      "mmlu_eval_accuracy_sociology": 0.7272727272727273,
-      "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
-      "mmlu_eval_accuracy_virology": 0.5,
+      "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": 0.8181818181818182,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
       "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
-      "mmlu_loss": 1.6376552364689252,
+      "mmlu_loss": 1.5687837348594367,
       "step": 400
     },
     {
-      "epoch": 0.16,
+      "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8306,
+      "loss": 0.9084,
       "step": 410
     },
     {
-      "epoch": 0.17,
+      "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7737,
+      "loss": 0.8347,
       "step": 420
     },
     {
-      "epoch": 0.17,
+      "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.7976,
+      "loss": 0.8994,
       "step": 430
     },
     {
-      "epoch": 0.18,
+      "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.846,
+      "loss": 0.8385,
       "step": 440
     },
     {
-      "epoch": 0.18,
+      "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.8067,
+      "loss": 0.7844,
       "step": 450
     },
     {
-      "epoch": 0.18,
+      "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.7816,
+      "loss": 0.7386,
       "step": 460
     },
     {
-      "epoch": 0.19,
+      "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.8445,
+      "loss": 0.8053,
       "step": 470
     },
     {
-      "epoch": 0.19,
+      "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.8055,
+      "loss": 0.8232,
       "step": 480
     },
     {
-      "epoch": 0.2,
+      "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.8062,
+      "loss": 0.784,
       "step": 490
     },
     {
-      "epoch": 0.2,
+      "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.8234,
+      "loss": 0.9117,
       "step": 500
     },
     {
-      "epoch": 0.2,
+      "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7704,
+      "loss": 0.8685,
       "step": 510
     },
     {
-      "epoch": 0.21,
+      "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7661,
+      "loss": 0.8039,
       "step": 520
     },
     {
-      "epoch": 0.21,
+      "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.7611,
+      "loss": 0.7753,
       "step": 530
     },
     {
-      "epoch": 0.22,
+      "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8551,
+      "loss": 0.8229,
       "step": 540
     },
     {
-      "epoch": 0.22,
+      "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8386,
+      "loss": 0.8525,
       "step": 550
     },
     {
-      "epoch": 0.22,
+      "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8192,
+      "loss": 0.8225,
       "step": 560
     },
     {
-      "epoch": 0.23,
+      "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.7847,
+      "loss": 0.8656,
       "step": 570
     },
     {
-      "epoch": 0.23,
+      "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8051,
+      "loss": 0.8633,
       "step": 580
     },
     {
-      "epoch": 0.24,
+      "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8711,
+      "loss": 0.7918,
       "step": 590
     },
     {
-      "epoch": 0.24,
+      "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8169,
+      "loss": 0.8414,
       "step": 600
     },
     {
-      "epoch": 0.24,
-      "eval_loss": 0.7934040427207947,
-      "eval_runtime": 137.9714,
-      "eval_samples_per_second": 7.248,
-      "eval_steps_per_second": 3.624,
+      "epoch": 0.12,
+      "eval_loss": 0.806236207485199,
+      "eval_runtime": 187.2336,
+      "eval_samples_per_second": 5.341,
+      "eval_steps_per_second": 2.67,
       "step": 600
     },
     {
-      "epoch": 0.24,
-      "mmlu_eval_accuracy": 0.5021762076506284,
-      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "epoch": 0.12,
+      "mmlu_eval_accuracy": 0.4972022603401932,
+      "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_astronomy": 0.4375,
+      "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": 0.25,
       "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": 0.45454545454545453,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
-      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
+      "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": 0.36585365853658536,
-      "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
-      "mmlu_eval_accuracy_global_facts": 0.4,
-      "mmlu_eval_accuracy_high_school_biology": 0.375,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
+      "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+      "mmlu_eval_accuracy_global_facts": 0.5,
+      "mmlu_eval_accuracy_high_school_biology": 0.40625,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.4090909090909091,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.6190476190476191,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
       "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
-      "mmlu_eval_accuracy_high_school_physics": 0.23529411764705882,
-      "mmlu_eval_accuracy_high_school_psychology": 0.85,
-      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
+      "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8333333333333334,
+      "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": 0.7391304347826086,
+      "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": 0.36363636363636365,
@@ -561,29 +561,29 @@
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
-      "mmlu_eval_accuracy_moral_disputes": 0.39473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.22,
-      "mmlu_eval_accuracy_nutrition": 0.696969696969697,
-      "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
-      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
-      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
-      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_moral_disputes": 0.42105263157894735,
+      "mmlu_eval_accuracy_moral_scenarios": 0.21,
+      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_prehistory": 0.5428571428571428,
+      "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
+      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
       "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_public_relations": 0.6666666666666666,
+      "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": 0.7894736842105263,
-      "mmlu_loss": 1.4682848513437625,
+      "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
+      "mmlu_eval_accuracy_virology": 0.5555555555555556,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.5916847097188935,
       "step": 600
     }
   ],
   "max_steps": 10000,
-  "num_train_epochs": 4,
-  "total_flos": 2.10893074184405e+17,
+  "num_train_epochs": 2,
+  "total_flos": 8.532996517153997e+16,
   "trial_name": null,
   "trial_params": null
 }

checkpoint-600/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f3ffe1e648921a382291ef39d8cc0cd1028979aed640da2c0f57ba92ef0d4c1
+oid sha256:3bbe31a14918402dc9f1acefdd42b0a5be81c28fa58723448f4a4c73673c2bea
 size 5819