Training in progress, step 200

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"
 }

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63c05fd4f91c9a5fc0ae6e49bbe471f127fe0d6cbcd4af84d5281b33686d9d9f
+oid sha256:ba29820f50a3aef4023bffc431908968574c18b0a0ef0d61ad48f7f7cb4bebf0
 size 319977229

checkpoint-200/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_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-200/adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d60fab2d14042a664439da5be95ae94b3e96879568c4f9e0ce36ea3e8ee48832
+oid sha256:ba29820f50a3aef4023bffc431908968574c18b0a0ef0d61ad48f7f7cb4bebf0
 size 319977229

checkpoint-200/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a19f3c58a12fb9be151229dee1b8b1d4b236ec5c88439da3d00c46972c1215d4
+oid sha256:380cdceff0ac0eaa9af0178834daf879f98fb6cf73fa80d2d1c23ef8dfbb108c
 size 1279539525

checkpoint-200/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c073aab62c86fb23480339b5babb629b26220dc7e918c39c849768810323c442
+oid sha256:6a03047afbbb96f48621a1a7ce89f5a09c0f425c5abeac7bac0a9efd8bf7f5a3
 size 14511

checkpoint-200/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 0.8124681711196899,
+  "best_metric": 0.823621928691864,
   "best_model_checkpoint": "experts/expert-21/checkpoint-200",
-  "epoch": 0.07984031936127745,
+  "epoch": 0.03992015968063872,
   "global_step": 200,
   "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,
@@ -173,35 +173,35 @@
       "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
     }
   ],
   "max_steps": 10000,
-  "num_train_epochs": 4,
-  "total_flos": 7.039326356235878e+16,
+  "num_train_epochs": 2,
+  "total_flos": 2.821303091596493e+16,
   "trial_name": null,
   "trial_params": null
 }

checkpoint-200/training_args.bin

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

training_args.bin

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