checkpoint-800 for llama2-13b-chat

README.md

@@ -8,7 +8,15 @@ license: llama2
 
 0. using the updated [Meta's LLaMA-2 models](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf).
 1. support [4-bit qlora](https://arxiv.org/abs/2305.14314), extreme GPU memory and inference time saving;
-2. comparable MMLU evaluation dataset results, llama2-13b's 54.8% to our 54.34% (-0.46%).
+2. comparable MMLU evaluation dataset results, llama2-13b-chat:
+
+ |               | eval   | test   | comp-eval | comp-test |
+ |---------------|--------|--------|-----------|-----------|
+ |llama2-13b-chat| 54.58% | -      |           |           |
+ |ckpt-800       | 53.86% | 53.32% | -0.72%    | -         |
+
+ llama2-13b-chat: "7389082e6bc4fcbf6202e6108a70194800e6c51e"
+
 3. This lazy-lora adapter is based on [Meta's LLaMA-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf), and using the [oasst1 dataset](https://huggingface.co/datasets/OpenAssistant/oasst1), following [Guanaco](https://huggingface.co/timdettmers/guanaco-65b).
 
 ### Introduction
@@ -84,67 +92,131 @@ model.print_trainable_parameters()
 
 ## MMLU result:
 
+### MMLU eval result:
+
 ```json
-{"mmlu_loss": 1.818716175022038,
- "mmlu_eval_accuracy_public_relations": 0.6666666666666666,
- "mmlu_eval_accuracy_college_medicine": 0.36363636363636365,
+{"mmlu_loss": 1.6594436656097273,
+ "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
+ "mmlu_eval_accuracy_high_school_biology": 0.5,
+ "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+ "mmlu_eval_accuracy_jurisprudence": 0.6363636363636364,
+ "mmlu_eval_accuracy_virology": 0.4444444444444444,
+ "mmlu_eval_accuracy_logical_fallacies": 0.6666666666666666,
+ "mmlu_eval_accuracy_professional_law": 0.3176470588235294,
+ "mmlu_eval_accuracy_econometrics": 0.3333333333333333,
  "mmlu_eval_accuracy_prehistory": 0.6571428571428571,
- "mmlu_eval_accuracy_astronomy": 0.5625,
- "mmlu_eval_accuracy_moral_scenarios": 0.27,
- "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
- "mmlu_eval_accuracy_security_studies": 0.6296296296296297,
- "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
- "mmlu_eval_accuracy_sociology": 0.7272727272727273,
- "mmlu_eval_accuracy_professional_law": 0.3352941176470588,
- "mmlu_eval_accuracy_high_school_macroeconomics": 0.5348837209302325,
- "mmlu_eval_accuracy_college_chemistry": 0.375,
- "mmlu_eval_accuracy_us_foreign_policy": 0.7272727272727273,
- "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
- "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
- "mmlu_eval_accuracy_high_school_chemistry": 0.22727272727272727,
- "mmlu_eval_accuracy_electrical_engineering": 0.375,
- "mmlu_eval_accuracy_nutrition": 0.6666666666666666,
  "mmlu_eval_accuracy_professional_accounting": 0.41935483870967744,
- "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
- "mmlu_eval_accuracy_professional_medicine": 0.4838709677419355,
- "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
- "mmlu_eval_accuracy_miscellaneous": 0.686046511627907,
- "mmlu_eval_accuracy_virology": 0.4444444444444444,
- "mmlu_eval_accuracy_college_computer_science": 0.45454545454545453,
- "mmlu_eval_accuracy_international_law": 0.9230769230769231,
- "mmlu_eval_accuracy_logical_fallacies": 0.8333333333333334,
- "mmlu_eval_accuracy_high_school_biology": 0.5,
- "mmlu_eval_accuracy_abstract_algebra": 0.45454545454545453,
- "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
- "mmlu_eval_accuracy_high_school_microeconomics": 0.5384615384615384,
- "mmlu_eval_accuracy_medical_genetics": 0.7272727272727273,
- "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
- "mmlu_eval_accuracy_marketing": 0.76,
- "mmlu_eval_accuracy_human_sexuality": 0.5,
- "mmlu_eval_accuracy_econometrics": 0.4166666666666667,
- "mmlu_eval_accuracy_college_mathematics": 0.45454545454545453,
- "mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
- "mmlu_eval_accuracy_moral_disputes": 0.5526315789473685,
- "mmlu_eval_accuracy_high_school_geography": 0.7727272727272727,
+ "mmlu_eval_accuracy_professional_psychology": 0.4782608695652174,
  "mmlu_eval_accuracy_management": 0.8181818181818182,
+ "mmlu_eval_accuracy_human_sexuality": 0.5,
+ "mmlu_eval_accuracy_college_mathematics": 0.5454545454545454,
+ "mmlu_eval_accuracy_us_foreign_policy": 0.8181818181818182,
+ "mmlu_eval_accuracy_high_school_european_history": 0.6666666666666666,
+ "mmlu_eval_accuracy_miscellaneous": 0.7441860465116279,
+ "mmlu_eval_accuracy_international_law": 0.8461538461538461,
  "mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
- "mmlu_eval_accuracy_machine_learning": 0.2727272727272727,
- "mmlu_eval_accuracy_high_school_us_history": 0.8636363636363636,
- "mmlu_eval_accuracy_business_ethics": 0.45454545454545453,
+ "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
+ "mmlu_eval_accuracy_high_school_physics": 0.29411764705882354,
+ "mmlu_eval_accuracy_moral_scenarios": 0.28,
+ "mmlu_eval_accuracy_sociology": 0.7727272727272727,
+ "mmlu_eval_accuracy_professional_medicine": 0.45161290322580644,
  "mmlu_eval_accuracy_conceptual_physics": 0.4230769230769231,
- "mmlu_eval_accuracy_global_facts": 0.5,
- "mmlu_eval_accuracy_college_biology": 0.625,
- "mmlu_eval_accuracy_elementary_mathematics": 0.2682926829268293,
- "mmlu_eval_accuracy_high_school_world_history": 0.5769230769230769,
+ "mmlu_eval_accuracy_high_school_us_history": 0.8636363636363636,
+ "mmlu_eval_accuracy_clinical_knowledge": 0.5172413793103449,
  "mmlu_eval_accuracy_human_aging": 0.6086956521739131,
- "mmlu_eval_accuracy_jurisprudence": 0.6363636363636364,
- "mmlu_eval_accuracy_philosophy": 0.5,
- "mmlu_eval_accuracy_professional_psychology": 0.5217391304347826,
- "mmlu_eval_accuracy_world_religions": 0.7894736842105263,
- "mmlu_eval_accuracy_computer_security": 0.6363636363636364,
- "mmlu_eval_accuracy_high_school_psychology": 0.8,
- "mmlu_eval_accuracy": 0.5433557168763036,
- "epoch": 2.37}
+ "mmlu_eval_accuracy_college_medicine": 0.4090909090909091,
+ "mmlu_eval_accuracy_computer_security": 0.7272727272727273,
+ "mmlu_eval_accuracy_moral_disputes": 0.5789473684210527,
+ "mmlu_eval_accuracy_security_studies": 0.6296296296296297,
+ "mmlu_eval_accuracy_high_school_world_history": 0.5769230769230769,
+ "mmlu_eval_accuracy_public_relations": 0.5833333333333334,
+ "mmlu_eval_accuracy_medical_genetics": 0.7272727272727273,
+ "mmlu_eval_accuracy_electrical_engineering": 0.375,
+ "mmlu_eval_accuracy_marketing": 0.8,
+ "mmlu_eval_accuracy_high_school_geography": 0.7272727272727273,
+ "mmlu_eval_accuracy_high_school_government_and_politics": 0.5714285714285714,
+ "mmlu_eval_accuracy_abstract_algebra": 0.2727272727272727,
+ "mmlu_eval_accuracy_nutrition": 0.6363636363636364,
+ "mmlu_eval_accuracy_college_biology": 0.625,
+ "mmlu_eval_accuracy_formal_logic": 0.14285714285714285,
+ "mmlu_eval_accuracy_machine_learning": 0.5454545454545454,
+ "mmlu_eval_accuracy_high_school_psychology": 0.7333333333333333,
+ "mmlu_eval_accuracy_high_school_statistics": 0.34782608695652173,
+ "mmlu_eval_accuracy_philosophy": 0.5588235294117647,
+ "mmlu_eval_accuracy_high_school_microeconomics": 0.5769230769230769,
+ "mmlu_eval_accuracy_global_facts": 0.5,
+ "mmlu_eval_accuracy_anatomy": 0.6428571428571429,
+ "mmlu_eval_accuracy_college_computer_science": 0.36363636363636365,
+ "mmlu_eval_accuracy_college_physics": 0.5454545454545454,
+ "mmlu_eval_accuracy_high_school_chemistry": 0.2727272727272727,
+ "mmlu_eval_accuracy_astronomy": 0.5625,
+ "mmlu_eval_accuracy_elementary_mathematics": 0.21951219512195122,
+ "mmlu_eval_accuracy_high_school_macroeconomics": 0.4418604651162791,
+ "mmlu_eval_accuracy_college_chemistry": 0.25,
+ "mmlu_eval_accuracy": 0.5385831470660036}
+```
+
+### MMLU test result:
+```json
+{"mmlu_loss": 1.6477740873911495,
+ "mmlu_test_accuracy_us_foreign_policy": 0.76,
+ "mmlu_test_accuracy_conceptual_physics": 0.3659574468085106,
+ "mmlu_test_accuracy_professional_accounting": 0.38652482269503546,
+ "mmlu_test_accuracy_high_school_world_history": 0.7088607594936709,
+ "mmlu_test_accuracy_human_aging": 0.6547085201793722,
+ "mmlu_test_accuracy_clinical_knowledge": 0.569811320754717,
+ "mmlu_test_accuracy_abstract_algebra": 0.36,
+ "mmlu_test_accuracy_machine_learning": 0.3392857142857143,
+ "mmlu_test_accuracy_high_school_geography": 0.6767676767676768,
+ "mmlu_test_accuracy_medical_genetics": 0.54,
+ "mmlu_test_accuracy_virology": 0.4939759036144578,
+ "mmlu_test_accuracy_professional_medicine": 0.4889705882352941,
+ "mmlu_test_accuracy_philosophy": 0.594855305466238,
+ "mmlu_test_accuracy_logical_fallacies": 0.656441717791411,
+ "mmlu_test_accuracy_formal_logic": 0.2857142857142857,
+ "mmlu_test_accuracy_electrical_engineering": 0.5103448275862069,
+ "mmlu_test_accuracy_anatomy": 0.4962962962962963,
+ "mmlu_test_accuracy_computer_security": 0.68,
+ "mmlu_test_accuracy_high_school_physics": 0.3509933774834437,
+ "mmlu_test_accuracy_high_school_statistics": 0.37962962962962965,
+ "mmlu_test_accuracy_high_school_us_history": 0.7009803921568627,
+ "mmlu_test_accuracy_college_biology": 0.5347222222222222,
+ "mmlu_test_accuracy_college_mathematics": 0.32,
+ "mmlu_test_accuracy_marketing": 0.7606837606837606,
+ "mmlu_test_accuracy_moral_scenarios": 0.2849162011173184,
+ "mmlu_test_accuracy_high_school_mathematics": 0.3148148148148148,
+ "mmlu_test_accuracy_high_school_microeconomics": 0.5168067226890757,
+ "mmlu_test_accuracy_college_computer_science": 0.48,
+ "mmlu_test_accuracy_college_chemistry": 0.35,
+ "mmlu_test_accuracy_global_facts": 0.31,
+ "mmlu_test_accuracy_management": 0.6990291262135923,
+ "mmlu_test_accuracy_security_studies": 0.6204081632653061,
+ "mmlu_test_accuracy_high_school_psychology": 0.7211009174311926,
+ "mmlu_test_accuracy_international_law": 0.7272727272727273,
+ "mmlu_test_accuracy_college_medicine": 0.44508670520231214,
+ "mmlu_test_accuracy_professional_psychology": 0.5098039215686274,
+ "mmlu_test_accuracy_high_school_european_history": 0.6545454545454545,
+ "mmlu_test_accuracy_prehistory": 0.5925925925925926,
+ "mmlu_test_accuracy_business_ethics": 0.51,
+ "mmlu_test_accuracy_high_school_chemistry": 0.45320197044334976,
+ "mmlu_test_accuracy_high_school_government_and_politics": 0.7461139896373057,
+ "mmlu_test_accuracy_astronomy": 0.5723684210526315,
+ "mmlu_test_accuracy_human_sexuality": 0.5877862595419847,
+ "mmlu_test_accuracy_miscellaneous": 0.735632183908046,
+ "mmlu_test_accuracy_public_relations": 0.6181818181818182,
+ "mmlu_test_accuracy_elementary_mathematics": 0.35185185185185186,
+ "mmlu_test_accuracy_world_religions": 0.7602339181286549,
+ "mmlu_test_accuracy_moral_disputes": 0.5838150289017341,
+ "mmlu_test_accuracy_econometrics": 0.2894736842105263,
+ "mmlu_test_accuracy_high_school_computer_science": 0.58,
+ "mmlu_test_accuracy_jurisprudence": 0.6296296296296297,
+ "mmlu_test_accuracy_nutrition": 0.5980392156862745,
+ "mmlu_test_accuracy_high_school_macroeconomics": 0.4897435897435897,
+ "mmlu_test_accuracy_professional_law": 0.36962190352020863,
+ "mmlu_test_accuracy_high_school_biology": 0.635483870967742,
+ "mmlu_test_accuracy_college_physics": 0.3235294117647059,
+ "mmlu_test_accuracy_sociology": 0.7164179104477612,
+ "mmlu_test_accuracy": 0.5332109924946602}
 ```
 
 ## License and intended use 

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:523e17537dd552110194c3dbcb3d267265234c4ba50764c4aed1e4bd521ba685
+oid sha256:11500d8165c5ac7429525ca57ac156125403c9681a96924ab21f249262b61e6f
 size 500857293