Training in progress, step 800

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:316c05afc65201a82db1fe35558b74edc638e2bf7fc8b812ef6a0a8584071b89
+oid sha256:e76a12b40d96b039c516a755ff0fc82c6483336c390439b21ebc6df94896779f
 size 319977229

checkpoint-600/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-600/adapter_model/adapter_model/adapter_config.json

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

checkpoint-600/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adb9e9477e25e5f833e174c991243e847d63a5fecd0c370719b4465628a0deda
+oid sha256:316c05afc65201a82db1fe35558b74edc638e2bf7fc8b812ef6a0a8584071b89
 size 319977229

checkpoint-800/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-800/adapter_config.json

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

checkpoint-800/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48c98b0f1b54c6fc81b65004bc46e17f6849f7279191b69433376de49ee40588
+oid sha256:e76a12b40d96b039c516a755ff0fc82c6483336c390439b21ebc6df94896779f
 size 319977229

checkpoint-800/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37dc9c1b7f600c52fffdec4850c5e89dabdf9ee2ac15dd84ea365744e9a23683
+oid sha256:be25d67708276096612e3d4f1c60ff2d6079396ce302ff645fda3c9e333cb0f9
 size 1279539973

checkpoint-800/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71d6e45f0d667730d4d2f18005c58f35d563043cc958d6630f436fdad1f5953d
+oid sha256:8b2ac0bbb2415a408ba2161d5dc7b93cf5d07edaece22603a20a05406fb2f35f
 size 14511

checkpoint-800/trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.7561990022659302,
+  "best_metric": 0.7563537359237671,
   "best_model_checkpoint": "experts/expert-16/checkpoint-800",
   "epoch": 0.2534854245880862,
   "global_step": 800,
@@ -10,7 +10,7 @@
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.8331,
+      "loss": 0.8339,
       "step": 10
     },
     {
@@ -22,7 +22,7 @@
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.9038,
+      "loss": 0.9041,
       "step": 30
     },
     {
@@ -34,13 +34,13 @@
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8155,
+      "loss": 0.8151,
       "step": 50
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.7896,
+      "loss": 0.79,
       "step": 60
     },
     {
@@ -52,105 +52,105 @@
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8827,
+      "loss": 0.8831,
       "step": 80
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.861,
+      "loss": 0.8607,
       "step": 90
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.7879,
+      "loss": 0.7876,
       "step": 100
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.803,
+      "loss": 0.8031,
       "step": 110
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8212,
+      "loss": 0.8207,
       "step": 120
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8075,
+      "loss": 0.807,
       "step": 130
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.9263,
+      "loss": 0.9262,
       "step": 140
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.7969,
+      "loss": 0.7964,
       "step": 150
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.7883,
+      "loss": 0.7879,
       "step": 160
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.7582,
+      "loss": 0.7587,
       "step": 170
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8095,
+      "loss": 0.8091,
       "step": 180
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8614,
+      "loss": 0.8615,
       "step": 190
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8674,
+      "loss": 0.8672,
       "step": 200
     },
     {
       "epoch": 0.06,
-      "eval_loss": 0.7774001359939575,
-      "eval_runtime": 149.8878,
-      "eval_samples_per_second": 6.672,
-      "eval_steps_per_second": 3.336,
+      "eval_loss": 0.7779108881950378,
+      "eval_runtime": 110.9863,
+      "eval_samples_per_second": 9.01,
+      "eval_steps_per_second": 4.505,
       "step": 200
     },
     {
       "epoch": 0.06,
-      "mmlu_eval_accuracy": 0.4759538024775667,
+      "mmlu_eval_accuracy": 0.4744171116325413,
       "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
       "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4482758620689655,
       "mmlu_eval_accuracy_college_biology": 0.4375,
-      "mmlu_eval_accuracy_college_chemistry": 0.25,
+      "mmlu_eval_accuracy_college_chemistry": 0.125,
       "mmlu_eval_accuracy_college_computer_science": 0.18181818181818182,
       "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": 0.3076923076923077,
+      "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
       "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
       "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
@@ -159,16 +159,16 @@
       "mmlu_eval_accuracy_high_school_biology": 0.375,
       "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_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": 0.4418604651162791,
       "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.3076923076923077,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.391304347826087,
-      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
       "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
@@ -177,306 +177,306 @@
       "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": 0.6744186046511628,
+      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
       "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_moral_scenarios": 0.27,
       "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
       "mmlu_eval_accuracy_philosophy": 0.5,
       "mmlu_eval_accuracy_prehistory": 0.42857142857142855,
       "mmlu_eval_accuracy_professional_accounting": 0.3225806451612903,
-      "mmlu_eval_accuracy_professional_law": 0.3058823529411765,
+      "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
       "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
-      "mmlu_eval_accuracy_professional_psychology": 0.463768115942029,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
-      "mmlu_eval_accuracy_sociology": 0.6363636363636364,
+      "mmlu_eval_accuracy_sociology": 0.6818181818181818,
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5555555555555556,
-      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.619096915108728,
+      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
+      "mmlu_loss": 1.5868234255450824,
       "step": 200
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8312,
+      "loss": 0.8316,
       "step": 210
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8465,
+      "loss": 0.8454,
       "step": 220
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8433,
+      "loss": 0.8434,
       "step": 230
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8223,
+      "loss": 0.821,
       "step": 240
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7884,
+      "loss": 0.7893,
       "step": 250
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8233,
+      "loss": 0.8242,
       "step": 260
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.812,
+      "loss": 0.8128,
       "step": 270
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.8342,
+      "loss": 0.8344,
       "step": 280
     },
     {
       "epoch": 0.09,
       "learning_rate": 0.0002,
-      "loss": 0.8316,
+      "loss": 0.8338,
       "step": 290
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7984,
+      "loss": 0.7981,
       "step": 300
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7821,
+      "loss": 0.781,
       "step": 310
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7715,
+      "loss": 0.7717,
       "step": 320
     },
     {
       "epoch": 0.1,
       "learning_rate": 0.0002,
-      "loss": 0.7675,
+      "loss": 0.767,
       "step": 330
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.793,
+      "loss": 0.7925,
       "step": 340
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.8223,
+      "loss": 0.8226,
       "step": 350
     },
     {
       "epoch": 0.11,
       "learning_rate": 0.0002,
-      "loss": 0.7916,
+      "loss": 0.7912,
       "step": 360
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.8094,
+      "loss": 0.8093,
       "step": 370
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.7655,
+      "loss": 0.7648,
       "step": 380
     },
     {
       "epoch": 0.12,
       "learning_rate": 0.0002,
-      "loss": 0.7868,
+      "loss": 0.7866,
       "step": 390
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.7983,
+      "loss": 0.7976,
       "step": 400
     },
     {
       "epoch": 0.13,
-      "eval_loss": 0.7662714719772339,
-      "eval_runtime": 149.8239,
-      "eval_samples_per_second": 6.675,
-      "eval_steps_per_second": 3.337,
+      "eval_loss": 0.7656086683273315,
+      "eval_runtime": 110.9802,
+      "eval_samples_per_second": 9.011,
+      "eval_steps_per_second": 4.505,
       "step": 400
     },
     {
       "epoch": 0.13,
-      "mmlu_eval_accuracy": 0.47563130411411997,
+      "mmlu_eval_accuracy": 0.47124130233512024,
       "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": 0.4827586206896552,
+      "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": 0.18181818181818182,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
       "mmlu_eval_accuracy_college_medicine": 0.45454545454545453,
-      "mmlu_eval_accuracy_college_physics": 0.45454545454545453,
-      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
+      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
       "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
       "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.3170731707317073,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
       "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
-      "mmlu_eval_accuracy_global_facts": 0.3,
+      "mmlu_eval_accuracy_global_facts": 0.4,
       "mmlu_eval_accuracy_high_school_biology": 0.40625,
       "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+      "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": 0.46511627906976744,
       "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.34615384615384615,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
-      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
-      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
+      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
       "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
-      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
+      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
       "mmlu_eval_accuracy_logical_fallacies": 0.6111111111111112,
       "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
       "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.24,
-      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+      "mmlu_eval_accuracy_moral_scenarios": 0.25,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
       "mmlu_eval_accuracy_philosophy": 0.5,
-      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
       "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
       "mmlu_eval_accuracy_professional_law": 0.3058823529411765,
       "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
       "mmlu_eval_accuracy_professional_psychology": 0.5217391304347826,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
-      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "mmlu_eval_accuracy_security_studies": 0.5185185185185185,
       "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": 1.474700499685875,
+      "mmlu_loss": 1.4339068503199297,
       "step": 400
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.8181,
+      "loss": 0.8182,
       "step": 410
     },
     {
       "epoch": 0.13,
       "learning_rate": 0.0002,
-      "loss": 0.8442,
+      "loss": 0.8438,
       "step": 420
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.8194,
+      "loss": 0.8184,
       "step": 430
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.8192,
+      "loss": 0.8202,
       "step": 440
     },
     {
       "epoch": 0.14,
       "learning_rate": 0.0002,
-      "loss": 0.8265,
+      "loss": 0.8264,
       "step": 450
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.8383,
+      "loss": 0.8384,
       "step": 460
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.8375,
+      "loss": 0.8372,
       "step": 470
     },
     {
       "epoch": 0.15,
       "learning_rate": 0.0002,
-      "loss": 0.808,
+      "loss": 0.8072,
       "step": 480
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.8209,
+      "loss": 0.8214,
       "step": 490
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.8144,
+      "loss": 0.814,
       "step": 500
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.8465,
+      "loss": 0.847,
       "step": 510
     },
     {
       "epoch": 0.16,
       "learning_rate": 0.0002,
-      "loss": 0.8437,
+      "loss": 0.8444,
       "step": 520
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.8091,
+      "loss": 0.8096,
       "step": 530
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.8501,
+      "loss": 0.8496,
       "step": 540
     },
     {
       "epoch": 0.17,
       "learning_rate": 0.0002,
-      "loss": 0.7731,
+      "loss": 0.7729,
       "step": 550
     },
     {
@@ -488,19 +488,19 @@
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.749,
+      "loss": 0.7478,
       "step": 570
     },
     {
       "epoch": 0.18,
       "learning_rate": 0.0002,
-      "loss": 0.7947,
+      "loss": 0.7953,
       "step": 580
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.736,
+      "loss": 0.7363,
       "step": 590
     },
     {
@@ -511,47 +511,47 @@
     },
     {
       "epoch": 0.19,
-      "eval_loss": 0.7616425156593323,
-      "eval_runtime": 149.4328,
-      "eval_samples_per_second": 6.692,
-      "eval_steps_per_second": 3.346,
+      "eval_loss": 0.7616064548492432,
+      "eval_runtime": 110.9404,
+      "eval_samples_per_second": 9.014,
+      "eval_steps_per_second": 4.507,
       "step": 600
     },
     {
       "epoch": 0.19,
-      "mmlu_eval_accuracy": 0.4814898904813968,
+      "mmlu_eval_accuracy": 0.4749850916074463,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
       "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
       "mmlu_eval_accuracy_astronomy": 0.4375,
       "mmlu_eval_accuracy_business_ethics": 0.5454545454545454,
-      "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
+      "mmlu_eval_accuracy_clinical_knowledge": 0.4827586206896552,
       "mmlu_eval_accuracy_college_biology": 0.4375,
       "mmlu_eval_accuracy_college_chemistry": 0.25,
-      "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
+      "mmlu_eval_accuracy_college_computer_science": 0.18181818181818182,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
-      "mmlu_eval_accuracy_college_medicine": 0.3181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.2727272727272727,
       "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
       "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
-      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
       "mmlu_eval_accuracy_elementary_mathematics": 0.2682926829268293,
       "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
-      "mmlu_eval_accuracy_global_facts": 0.4,
+      "mmlu_eval_accuracy_global_facts": 0.3,
       "mmlu_eval_accuracy_high_school_biology": 0.40625,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.3181818181818182,
+      "mmlu_eval_accuracy_high_school_chemistry": 0.36363636363636365,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.3076923076923077,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
       "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.43478260869565216,
+      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
       "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
-      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_high_school_world_history": 0.7692307692307693,
       "mmlu_eval_accuracy_human_aging": 0.6521739130434783,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
@@ -559,16 +559,16 @@
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
       "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
-      "mmlu_eval_accuracy_marketing": 0.88,
+      "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
+      "mmlu_eval_accuracy_miscellaneous": 0.6511627906976745,
       "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.27,
+      "mmlu_eval_accuracy_moral_scenarios": 0.26,
       "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
-      "mmlu_eval_accuracy_philosophy": 0.47058823529411764,
-      "mmlu_eval_accuracy_prehistory": 0.4857142857142857,
-      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
-      "mmlu_eval_accuracy_professional_law": 0.3058823529411765,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
+      "mmlu_eval_accuracy_prehistory": 0.5142857142857142,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.3,
       "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
       "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
@@ -577,61 +577,61 @@
       "mmlu_eval_accuracy_us_foreign_policy": 0.6363636363636364,
       "mmlu_eval_accuracy_virology": 0.5,
       "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
-      "mmlu_loss": 1.5487773687658983,
+      "mmlu_loss": 1.5647042619341658,
       "step": 600
     },
     {
       "epoch": 0.19,
       "learning_rate": 0.0002,
-      "loss": 0.7917,
+      "loss": 0.7936,
       "step": 610
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.7324,
+      "loss": 0.7319,
       "step": 620
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.7894,
+      "loss": 0.79,
       "step": 630
     },
     {
       "epoch": 0.2,
       "learning_rate": 0.0002,
-      "loss": 0.7818,
+      "loss": 0.7806,
       "step": 640
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.8839,
+      "loss": 0.8833,
       "step": 650
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.7716,
+      "loss": 0.7711,
       "step": 660
     },
     {
       "epoch": 0.21,
       "learning_rate": 0.0002,
-      "loss": 0.8243,
+      "loss": 0.8242,
       "step": 670
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.7961,
+      "loss": 0.7948,
       "step": 680
     },
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.742,
+      "loss": 0.7417,
       "step": 690
     },
     {
@@ -643,88 +643,88 @@
     {
       "epoch": 0.22,
       "learning_rate": 0.0002,
-      "loss": 0.814,
+      "loss": 0.8137,
       "step": 710
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8581,
+      "loss": 0.8568,
       "step": 720
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8023,
+      "loss": 0.802,
       "step": 730
     },
     {
       "epoch": 0.23,
       "learning_rate": 0.0002,
-      "loss": 0.8199,
+      "loss": 0.8202,
       "step": 740
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.807,
+      "loss": 0.8077,
       "step": 750
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.8133,
+      "loss": 0.814,
       "step": 760
     },
     {
       "epoch": 0.24,
       "learning_rate": 0.0002,
-      "loss": 0.7978,
+      "loss": 0.7971,
       "step": 770
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.7977,
+      "loss": 0.798,
       "step": 780
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.7802,
+      "loss": 0.7806,
       "step": 790
     },
     {
       "epoch": 0.25,
       "learning_rate": 0.0002,
-      "loss": 0.8041,
+      "loss": 0.8042,
       "step": 800
     },
     {
       "epoch": 0.25,
-      "eval_loss": 0.7561990022659302,
-      "eval_runtime": 149.4612,
-      "eval_samples_per_second": 6.691,
-      "eval_steps_per_second": 3.345,
+      "eval_loss": 0.7563537359237671,
+      "eval_runtime": 111.023,
+      "eval_samples_per_second": 9.007,
+      "eval_steps_per_second": 4.504,
       "step": 800
     },
     {
       "epoch": 0.25,
-      "mmlu_eval_accuracy": 0.4774734261484596,
+      "mmlu_eval_accuracy": 0.4796267144005645,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
       "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
       "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": 0.5,
+      "mmlu_eval_accuracy_college_biology": 0.4375,
       "mmlu_eval_accuracy_college_chemistry": 0.125,
       "mmlu_eval_accuracy_college_computer_science": 0.2727272727272727,
       "mmlu_eval_accuracy_college_mathematics": 0.2727272727272727,
-      "mmlu_eval_accuracy_college_medicine": 0.3181818181818182,
+      "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
       "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
       "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
       "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
-      "mmlu_eval_accuracy_econometrics": 0.25,
+      "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
       "mmlu_eval_accuracy_elementary_mathematics": 0.2926829268292683,
       "mmlu_eval_accuracy_formal_logic": 0.07142857142857142,
@@ -732,43 +732,43 @@
       "mmlu_eval_accuracy_high_school_biology": 0.375,
       "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
       "mmlu_eval_accuracy_high_school_computer_science": 0.5555555555555556,
-      "mmlu_eval_accuracy_high_school_european_history": 0.6111111111111112,
+      "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.46511627906976744,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.38461538461538464,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
       "mmlu_eval_accuracy_high_school_psychology": 0.8833333333333333,
-      "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+      "mmlu_eval_accuracy_high_school_statistics": 0.2608695652173913,
       "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
       "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
-      "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
+      "mmlu_eval_accuracy_human_aging": 0.7391304347826086,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.8461538461538461,
-      "mmlu_eval_accuracy_jurisprudence": 0.36363636363636365,
+      "mmlu_eval_accuracy_jurisprudence": 0.45454545454545453,
       "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
-      "mmlu_eval_accuracy_machine_learning": 0.18181818181818182,
+      "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
       "mmlu_eval_accuracy_marketing": 0.84,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
-      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
-      "mmlu_eval_accuracy_moral_scenarios": 0.24,
+      "mmlu_eval_accuracy_miscellaneous": 0.6744186046511628,
+      "mmlu_eval_accuracy_moral_disputes": 0.5,
+      "mmlu_eval_accuracy_moral_scenarios": 0.23,
       "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
       "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
-      "mmlu_eval_accuracy_prehistory": 0.42857142857142855,
-      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
-      "mmlu_eval_accuracy_professional_law": 0.31176470588235294,
+      "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
+      "mmlu_eval_accuracy_professional_accounting": 0.3548387096774194,
+      "mmlu_eval_accuracy_professional_law": 0.3,
       "mmlu_eval_accuracy_professional_medicine": 0.5806451612903226,
-      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
+      "mmlu_eval_accuracy_professional_psychology": 0.5072463768115942,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.4074074074074074,
       "mmlu_eval_accuracy_sociology": 0.6363636363636364,
       "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
       "mmlu_eval_accuracy_virology": 0.5555555555555556,
-      "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
-      "mmlu_loss": 1.498360338394698,
+      "mmlu_eval_accuracy_world_religions": 0.6842105263157895,
+      "mmlu_loss": 1.4866046660796157,
       "step": 800
     }
   ],

checkpoint-800/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d55ea0037adc27e709c2f87c727ac3fd55b607c5afed1c76d4262111f780937e
+oid sha256:277dab2895cbf8d8b713e5f0339719c5d50d193c9afc1399ad953aa4910a2fab
 size 5819