GitHub - RealmX1/LLM-FineTuning-Confidence-Encoding-MSML641FinalProject

create virtual environment, perform pip install -r requirements.txt install LM Studio and download the respective models for each model

edit model_name in lm_studio_api.py, start server in LM Studio and run s1~s3; ADJUST PROMPT FOR PAIRWISE CONFIDENCE PHRASE COMPARISON ACCORDING TO EACH MODEL
fine-tune model in s4_unsloth_fine_tuning.ipynb -- it should be run in colab, and you should upload the synthetic_knowledge.csv to the colab. ADJUST PROMPT FOR FINETUNING ACCORDING TO EACH MODEL's MODEL CARD ON HUGGING FACE & other available sources!
save and download the fine-tuned model to be hosted in LM Studio (optional, but if you do it on colab alone you might reach time-limit for colab free time)
collect model's answer to the domain comparison question, and fill a domain_comparison_{model_name}.csv table
add the table to s5_evaluation.py, run it to see the analysis

Name		Name	Last commit message	Last commit date
Latest commit History 8 Commits
__pycache__		__pycache__
data		data
641 Final Project Presentation.pdf		641 Final Project Presentation.pdf
641 Final Project Presentation.pptx		641 Final Project Presentation.pptx
MSML641FinalReport.docx		MSML641FinalReport.docx
MSML641FinalReport.pdf		MSML641FinalReport.pdf
README.md		README.md
llm_reading_comprehension.csv		llm_reading_comprehension.csv
lm_studio_api.py		lm_studio_api.py
phrase_confidence_ranking.csv		phrase_confidence_ranking.csv
phrase_confidence_ranking_phi-3.csv		phrase_confidence_ranking_phi-3.csv
random_domains.txt		random_domains.txt
report.ipynb		report.ipynb
requirements.txt		requirements.txt
s1_pairwise_confidence_comparison.py		s1_pairwise_confidence_comparison.py
s2_analyze_comparison_result.py		s2_analyze_comparison_result.py
s3_synthetic_knowledge_factory.py		s3_synthetic_knowledge_factory.py
s4_unsloth_fine_tuning.ipynb		s4_unsloth_fine_tuning.ipynb
s5_2_evaluation_manual.py		s5_2_evaluation_manual.py
s5_evaluation.py		s5_evaluation.py
test.py		test.py

Provide feedback