# To create model file from output of LexFineTune do the following:
To run LoRA trained model in ollama:
1.) Run ModelMerge.py
2.) python ~/Projects/llama/convertToGGUF/convertToGGUF/llama.cpp/convert_hf_to_gguf.py /Users/csimoes/Projects/llama/final_Mar052024 --outfile lex_llama_unquantized.gguf --verbose --outtype f16
python /home/ubuntu/projects/llama.cpp/convert_hf_to_gguf.py ~/projects/LexLLM/lex_lora_results_20250307_162307/checkpoint_epoch_5 --outfile lex_llama_unquantized.gguf --verbose --outtype f16
2b.) [Optional] Test GGUF file manually: python ~/Projects/llama/convertToGGUF/convertToGGUF/llama.cpp/llama-cli -m lex_llama_unquantized.gguf -p "Tell me about AI."
2c.) [Optional] Quantize to reduce memory needs: llama.cpp/quantize lex_llama_unquantized.gguf lex_llama_q4_k_m.gguf Q4_K_M

3.) ollama create [model name] -f [modelfile]


# To watch performance of AWS machine
nvidia-smi  # Check GPU usage
htop        # Check CPU usage (should see 16 threads)

# NVidia GPU usage:
nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total --format=csv -l 5