Различия

Показаны различия между двумя версиями страницы.

--- it:nejrosetevaja_shpargalka [2025/07/15 07:09] – создано puse_vivat
+++ it:nejrosetevaja_shpargalka [2025/07/15 07:47] (текущий) – puse_vivat
@@ Строка 3: / Строка 3: @@
 ===== Сборка llama.cpp =====
-cd devel
+cd devel\\
-git clone https://github.com/ggml-org/llama.cpp
+git clone https://github.com/ggml-org/llama.cpp\\
-cd llama.cpp
+cd llama.cpp\\
-git pull
+git pull\\
-git fetch
+git fetch\\
-cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1102 -DCMAKE_BUILD_TYPE=Release
+cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=gfx1102 -DCMAKE_BUILD_TYPE=Release\\
-cmake --build build -j 24 --config Release
+cmake --build build -j 24 --config Release\\
-sudo cp build/bin/* /usr/local/bin/
+sudo cp build/bin/* /usr/local/bin/\\
+===== Конвертация модели в gguf =====
+~/devel/llama.cpp/convert_hf_to_gguf.py --outtype bf16 --outfile ~/ggufs/F/Qwen3-55B-A3B-TOTAL-RECALL-V1.3-128k-BF16.gguf ~/models/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-V1.3-128k/
+~/devel/llama.cpp/convert_hf_to_gguf.py --outfile ~/ggufs/VL/Qwen2.5-VL-7B-Instruct-BF16.gguf --outtype bf16 ~/models/Qwen/Qwen2.5-VL-7B-Instruct/\\
+~/devel/llama.cpp/convert_hf_to_gguf.py --outfile ~/ggufs/VL/Qwen2.5-VL-7B-Instruct-BF16.gguf --outtype bf16 --mmproj ~/models/Qwen/Qwen2.5-VL-7B-Instruct/
+===== Квантизация =====
+~/devel/llama.cpp/build/bin/llama-quantize ~/ggufs/F/DS-R1-0528-Qwen3-8B-BF16.gguf ~/ggufs/DS-R1-0528-Qwen3-8B-Q6_K.gguf Q6_K
 ===== Сервер с моделью text-to-text =====