Instructions to use VladHong/llama-cpp-K2-FAST2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use VladHong/llama-cpp-K2-FAST2 with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: ./build/bin/llama cli -hf VladHong/llama-cpp-K2-FAST2
Use Docker
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Llama new
- LM Studio
- Jan
- Ollama
How to use VladHong/llama-cpp-K2-FAST2 with Ollama:
ollama run hf.co/VladHong/llama-cpp-K2-FAST2
- Unsloth Desktop
- Docker Model Runner
How to use VladHong/llama-cpp-K2-FAST2 with Docker Model Runner:
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Lemonade
How to use VladHong/llama-cpp-K2-FAST2 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull VladHong/llama-cpp-K2-FAST2
Run and chat with the model
lemonade run user.llama-cpp-K2-FAST2-{{QUANT_TAG}}List all available models
lemonade list
- Atomic Chat
Download ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp from VladHong/llama-cpp-K2-FAST2: direct link, hf CLI and curl.
- Browser
- Download file 1.43 kB
-
https://huggingface.co/VladHong/llama-cpp-K2-FAST2/resolve/main/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp
- Command line
-
hf download hf://VladHong/llama-cpp-K2-FAST2/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp
-
curl -L -o copy_from_quant.comp https://huggingface.co/VladHong/llama-cpp-K2-FAST2/resolve/main/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp
1.43 kB
| // 16 invocations needed for init_iq_shmem | |
| layout(local_size_x = 16, local_size_y = 1, local_size_z = 1) in; | |
| layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in; | |
| void main() { | |
| init_iq_shmem(gl_WorkGroupSize); | |
| if (gl_LocalInvocationIndex.x != 0) { | |
| return; | |
| } | |
| const uint idx = gl_WorkGroupID.z * 262144 + gl_WorkGroupID.y * 512 + gl_WorkGroupID.x * QUANT_K; | |
| if (idx >= p.ne) { | |
| return; | |
| } | |
| uint dst_idx = get_doffset() + dst_idx(idx); | |
| uint src_idx = src0_idx_quant(idx, QUANT_K); | |
| const uint a_offset = 0; | |
| const uint ib = src_idx; | |
| const vec2 dm = get_dm(ib, a_offset); | |
| [[unroll]] for (int j = 0; j < QUANT_K; j += 4) { | |
| vec4 v = dequantize4(ib, j / QUANT_R, a_offset); | |
| v = v * dm.x + vec4(dm.y); | |
| data_d[dst_idx + j/2 + 0] = v[0]; | |
| data_d[dst_idx + j/2 + QUANT_K/2 + 0] = v[1]; | |
| data_d[dst_idx + j/2 + 1] = v[2]; | |
| data_d[dst_idx + j/2 + QUANT_K/2 + 1] = v[3]; | |
| data_d[dst_idx + j + 0] = v[0]; | |
| data_d[dst_idx + j + 1] = v[1]; | |
| data_d[dst_idx + j + 2] = v[2]; | |
| data_d[dst_idx + j + 3] = v[3]; | |
| } | |
| } | |