# Clone the repository
git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp

# Create build directory
mkdir build
cd build

# Configure for CUDA 13.0 and A100 (sm_80)
# We use -DGGML_CUDA=ON to enable GPU acceleration
cmake .. -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=80

# Build the binaries (using all CPU cores)
cmake --build . --config Release -j $(nproc)


cd ~/llama.cpp/build
rm -rf *

CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=2 ~/llama.cpp/build/bin/llama-server \
  -m /home/mshahidul/readctrl/models/translategemma-27b-it-Q8_0.gguf \
  --n-gpu-layers 999 \
  --flash-attn



  ~/llama.cpp/build/bin/llama-server \
  -m ~/models/Meta-Llama-3.1-70B-Instruct-Q4_K_M.gguf \
  -ngl 99 \
  -fa \
  --port 8080 \
  --host 0.0.0.0


python /home/mshahidul/readctrl/script/translate_correction_gpt5.py --input "/home/mshahidul/readctrl/data/translated_data/translation_wo_judge/multiclinsum_gs_train_en2zh_gemma(0_200).json" --start 0 --end 80

python /home/mshahidul/readctrl/script/translate_correction_gpt5.py --input "/home/mshahidul/readctrl/data/translated_data/translation_wo_judge/multiclinsum_gs_train_en2vi_gemma(0_200).json" --start 0 --end 80

python /home/mshahidul/readctrl/script/translate_correction_gpt5.py --input "/home/mshahidul/readctrl/data/translated_data/translation_wo_judge/multiclinsum_gs_train_en2hi_gemma(0_200).json" --start 0 --end 80