gguf.txt
513 B · 15 lines · plaintext Raw
1 # 1. Конвертация в bf16 (как договорились)
2 python convert_hf_to_gguf.py /mnt/nfs_share/JiRackUlrta_1 \
3 --outfile /mnt/nfs_share/JiRackUlrta_1/jirack_1p5b.gguf --outtype bf16
4
5 # 2. Сборка llama-quantize (один раз)
6 cd /mnt/nfs_share/llama.cpp
7 cmake -B build
8 cmake --build build -j
9
10 # 3. Квантование в Q4_K_M
11 ./build/bin/llama-quantize \
12 /mnt/nfs_share/JiRackUlrta_1/jirack_1p5b.gguf \
13 /mnt/nfs_share/JiRackUlrta_1/jirack_1p5b.Q4_K_M.gguf \
14 Q4_K_M
15