Training in progress, step 400

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2920cd03151b12944b05cac37aae55df3d589cf8838396f445986adf54625c8
+oid sha256:972cd2ebd9a6737a22febb70d1db40a67574598e21722edea4d4610876e195ba
 size 319977229

checkpoint-200/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-200/adapter_model/adapter_model/adapter_config.json

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

checkpoint-200/adapter_model/adapter_model/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:263033e4bde69e7abe4e2bb79382c8e341df3d121300b2764dd67d5250bcdf20
+oid sha256:e2920cd03151b12944b05cac37aae55df3d589cf8838396f445986adf54625c8
 size 319977229

checkpoint-400/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_config.json

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

checkpoint-400/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f1f11726260e5f15f3bfb21f6572f487cb0ac87d4816d145899ab0a4826eebe
+oid sha256:972cd2ebd9a6737a22febb70d1db40a67574598e21722edea4d4610876e195ba
 size 319977229

checkpoint-400/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1bf8e8d1f1611fcd4b30e4346f672daadb34b0ac5919c86eb670ccfe1d6b9658
+oid sha256:9efe9624cfaf1ce05e0b937857e92b71e2642749fbb6ed9088581ffa4383d7cb
 size 1279539973

checkpoint-400/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd6ae8059ff6f79d974532cf7d03ddffb8eb00829c82144ac71009bd9504a05c
+oid sha256:97371ecb09dd895a5aa7dffb5ea740a86e94cca6bfcbc51836a7357f3769c83b
 size 14511

checkpoint-400/trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.7932400703430176,
+  "best_metric": 0.7937927842140198,
   "best_model_checkpoint": "experts/expert-13/checkpoint-400",
   "epoch": 0.08406893652795293,
   "global_step": 400,
@@ -10,85 +10,85 @@
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.9192,
+      "loss": 0.9187,
       "step": 10
     },
     {
       "epoch": 0.0,
       "learning_rate": 0.0002,
-      "loss": 0.8713,
+      "loss": 0.8703,
       "step": 20
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.859,
+      "loss": 0.8582,
       "step": 30
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8968,
+      "loss": 0.8973,
       "step": 40
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8191,
+      "loss": 0.8188,
       "step": 50
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.8115,
+      "loss": 0.8114,
       "step": 60
     },
     {
       "epoch": 0.01,
       "learning_rate": 0.0002,
-      "loss": 0.7799,
+      "loss": 0.7796,
       "step": 70
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8676,
+      "loss": 0.8671,
       "step": 80
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.887,
+      "loss": 0.8863,
       "step": 90
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8152,
+      "loss": 0.815,
       "step": 100
     },
     {
       "epoch": 0.02,
       "learning_rate": 0.0002,
-      "loss": 0.8251,
+      "loss": 0.8255,
       "step": 110
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8274,
+      "loss": 0.8272,
       "step": 120
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8151,
+      "loss": 0.8148,
       "step": 130
     },
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8547,
+      "loss": 0.8542,
       "step": 140
     },
     {
@@ -100,7 +100,7 @@
     {
       "epoch": 0.03,
       "learning_rate": 0.0002,
-      "loss": 0.8747,
+      "loss": 0.8743,
       "step": 160
     },
     {
@@ -112,37 +112,37 @@
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.7856,
+      "loss": 0.7863,
       "step": 180
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.839,
+      "loss": 0.8397,
       "step": 190
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.7512,
+      "loss": 0.7508,
       "step": 200
     },
     {
       "epoch": 0.04,
-      "eval_loss": 0.8084444403648376,
-      "eval_runtime": 91.3593,
-      "eval_samples_per_second": 10.946,
-      "eval_steps_per_second": 5.473,
+      "eval_loss": 0.8083679676055908,
+      "eval_runtime": 68.466,
+      "eval_samples_per_second": 14.606,
+      "eval_steps_per_second": 7.303,
       "step": 200
     },
     {
       "epoch": 0.04,
-      "mmlu_eval_accuracy": 0.4842637889631141,
+      "mmlu_eval_accuracy": 0.4843923123312093,
       "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
-      "mmlu_eval_accuracy_anatomy": 0.7142857142857143,
+      "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_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,
@@ -153,61 +153,61 @@
       "mmlu_eval_accuracy_conceptual_physics": 0.34615384615384615,
       "mmlu_eval_accuracy_econometrics": 0.25,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
+      "mmlu_eval_accuracy_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": 0.2727272727272727,
       "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": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
-      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
-      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4883720930232558,
+      "mmlu_eval_accuracy_high_school_government_and_politics": 0.5238095238095238,
+      "mmlu_eval_accuracy_high_school_macroeconomics": 0.4186046511627907,
       "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
       "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
       "mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
       "mmlu_eval_accuracy_high_school_psychology": 0.85,
       "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
-      "mmlu_eval_accuracy_high_school_us_history": 0.5909090909090909,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
       "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
       "mmlu_eval_accuracy_human_aging": 0.6956521739130435,
       "mmlu_eval_accuracy_human_sexuality": 0.3333333333333333,
       "mmlu_eval_accuracy_international_law": 0.9230769230769231,
-      "mmlu_eval_accuracy_jurisprudence": 0.5454545454545454,
+      "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": 0.92,
+      "mmlu_eval_accuracy_marketing": 0.88,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
-      "mmlu_eval_accuracy_miscellaneous": 0.6976744186046512,
+      "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
       "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.23,
+      "mmlu_eval_accuracy_moral_scenarios": 0.24,
       "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
       "mmlu_eval_accuracy_philosophy": 0.4411764705882353,
       "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
-      "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
-      "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
+      "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
+      "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
       "mmlu_eval_accuracy_professional_medicine": 0.6129032258064516,
-      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
       "mmlu_eval_accuracy_security_studies": 0.4444444444444444,
       "mmlu_eval_accuracy_sociology": 0.7272727272727273,
       "mmlu_eval_accuracy_us_foreign_policy": 0.5454545454545454,
       "mmlu_eval_accuracy_virology": 0.4444444444444444,
       "mmlu_eval_accuracy_world_religions": 0.7368421052631579,
-      "mmlu_loss": 1.5038623624619865,
+      "mmlu_loss": 1.48628691881506,
       "step": 200
     },
     {
       "epoch": 0.04,
       "learning_rate": 0.0002,
-      "loss": 0.8045,
+      "loss": 0.8043,
       "step": 210
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8066,
+      "loss": 0.8064,
       "step": 220
     },
     {
@@ -219,116 +219,116 @@
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8211,
+      "loss": 0.8212,
       "step": 240
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.8053,
+      "loss": 0.8044,
       "step": 250
     },
     {
       "epoch": 0.05,
       "learning_rate": 0.0002,
-      "loss": 0.787,
+      "loss": 0.7862,
       "step": 260
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8109,
+      "loss": 0.8106,
       "step": 270
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.7959,
+      "loss": 0.7954,
       "step": 280
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8155,
+      "loss": 0.8153,
       "step": 290
     },
     {
       "epoch": 0.06,
       "learning_rate": 0.0002,
-      "loss": 0.8232,
+      "loss": 0.8235,
       "step": 300
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8077,
+      "loss": 0.8087,
       "step": 310
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.7821,
+      "loss": 0.7824,
       "step": 320
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8201,
+      "loss": 0.8198,
       "step": 330
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.8287,
+      "loss": 0.8288,
       "step": 340
     },
     {
       "epoch": 0.07,
       "learning_rate": 0.0002,
-      "loss": 0.7965,
+      "loss": 0.797,
       "step": 350
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7597,
+      "loss": 0.7599,
       "step": 360
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8158,
+      "loss": 0.8156,
       "step": 370
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.8537,
+      "loss": 0.8534,
       "step": 380
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7705,
+      "loss": 0.7712,
       "step": 390
     },
     {
       "epoch": 0.08,
       "learning_rate": 0.0002,
-      "loss": 0.7556,
+      "loss": 0.7555,
       "step": 400
     },
     {
       "epoch": 0.08,
-      "eval_loss": 0.7932400703430176,
-      "eval_runtime": 104.0519,
-      "eval_samples_per_second": 9.611,
-      "eval_steps_per_second": 4.805,
+      "eval_loss": 0.7937927842140198,
+      "eval_runtime": 68.5823,
+      "eval_samples_per_second": 14.581,
+      "eval_steps_per_second": 7.291,
       "step": 400
     },
     {
       "epoch": 0.08,
-      "mmlu_eval_accuracy": 0.4874623111988614,
+      "mmlu_eval_accuracy": 0.4820017142234064,
       "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
       "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
       "mmlu_eval_accuracy_astronomy": 0.4375,
@@ -338,55 +338,55 @@
       "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_medicine": 0.4090909090909091,
       "mmlu_eval_accuracy_college_physics": 0.36363636363636365,
-      "mmlu_eval_accuracy_computer_security": 0.45454545454545453,
-      "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
+      "mmlu_eval_accuracy_computer_security": 0.36363636363636365,
+      "mmlu_eval_accuracy_conceptual_physics": 0.38461538461538464,
       "mmlu_eval_accuracy_econometrics": 0.16666666666666666,
       "mmlu_eval_accuracy_electrical_engineering": 0.25,
-      "mmlu_eval_accuracy_elementary_mathematics": 0.36585365853658536,
+      "mmlu_eval_accuracy_elementary_mathematics": 0.34146341463414637,
       "mmlu_eval_accuracy_formal_logic": 0.21428571428571427,
       "mmlu_eval_accuracy_global_facts": 0.6,
-      "mmlu_eval_accuracy_high_school_biology": 0.46875,
-      "mmlu_eval_accuracy_high_school_chemistry": 0.18181818181818182,
+      "mmlu_eval_accuracy_high_school_biology": 0.4375,
+      "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": 0.5555555555555556,
       "mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
       "mmlu_eval_accuracy_high_school_government_and_politics": 0.47619047619047616,
       "mmlu_eval_accuracy_high_school_macroeconomics": 0.5116279069767442,
-      "mmlu_eval_accuracy_high_school_mathematics": 0.27586206896551724,
-      "mmlu_eval_accuracy_high_school_microeconomics": 0.4230769230769231,
+      "mmlu_eval_accuracy_high_school_mathematics": 0.20689655172413793,
+      "mmlu_eval_accuracy_high_school_microeconomics": 0.46153846153846156,
       "mmlu_eval_accuracy_high_school_physics": 0.17647058823529413,
-      "mmlu_eval_accuracy_high_school_psychology": 0.8666666666666667,
-      "mmlu_eval_accuracy_high_school_statistics": 0.30434782608695654,
-      "mmlu_eval_accuracy_high_school_us_history": 0.6818181818181818,
-      "mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
+      "mmlu_eval_accuracy_high_school_psychology": 0.85,
+      "mmlu_eval_accuracy_high_school_statistics": 0.2608695652173913,
+      "mmlu_eval_accuracy_high_school_us_history": 0.6363636363636364,
+      "mmlu_eval_accuracy_high_school_world_history": 0.6923076923076923,
       "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": 0.45454545454545453,
+      "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": 0.2727272727272727,
+      "mmlu_eval_accuracy_machine_learning": 0.36363636363636365,
       "mmlu_eval_accuracy_management": 0.6363636363636364,
-      "mmlu_eval_accuracy_marketing": 0.8,
+      "mmlu_eval_accuracy_marketing": 0.76,
       "mmlu_eval_accuracy_medical_genetics": 0.9090909090909091,
       "mmlu_eval_accuracy_miscellaneous": 0.6627906976744186,
-      "mmlu_eval_accuracy_moral_disputes": 0.4473684210526316,
-      "mmlu_eval_accuracy_moral_scenarios": 0.22,
-      "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
-      "mmlu_eval_accuracy_philosophy": 0.5,
+      "mmlu_eval_accuracy_moral_disputes": 0.47368421052631576,
+      "mmlu_eval_accuracy_moral_scenarios": 0.21,
+      "mmlu_eval_accuracy_nutrition": 0.6060606060606061,
+      "mmlu_eval_accuracy_philosophy": 0.5294117647058824,
       "mmlu_eval_accuracy_prehistory": 0.45714285714285713,
       "mmlu_eval_accuracy_professional_accounting": 0.3870967741935484,
       "mmlu_eval_accuracy_professional_law": 0.3411764705882353,
       "mmlu_eval_accuracy_professional_medicine": 0.5483870967741935,
-      "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
+      "mmlu_eval_accuracy_professional_psychology": 0.4927536231884058,
       "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
-      "mmlu_eval_accuracy_security_studies": 0.48148148148148145,
+      "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": 0.6842105263157895,
-      "mmlu_loss": 1.505364536071882,
+      "mmlu_loss": 1.500225733527938,
       "step": 400
     }
   ],

checkpoint-400/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c88cdffadce27d1fa15d3da3b2e83029e51e413f14a49ea322f0a4eee84d961
+oid sha256:fe7cdc1358006edc28e72abc36fb4c6703372887fc1df961096270e51aa9d994
 size 5819