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/dequant_q2_0.comp from VladHong/llama-cpp-K2-FAST2: direct link, hf CLI and curl.
- Browser
- Download file 938 Bytes
-
https://huggingface.co/VladHong/llama-cpp-K2-FAST2/resolve/main/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_0.comp
- Command line
-
hf download hf://VladHong/llama-cpp-K2-FAST2/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_0.comp
-
curl -L -o dequant_q2_0.comp https://huggingface.co/VladHong/llama-cpp-K2-FAST2/resolve/main/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_0.comp
938 Bytes
| layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in; | |
| layout (binding = 0) readonly buffer A {block_q2_0 data_a[];}; | |
| layout (binding = 1) writeonly buffer D {D_TYPE data_b[];}; | |
| void main() { | |
| const uint i = gl_WorkGroupID.x * 4 + gl_LocalInvocationID.x / 64; | |
| const uint tid = gl_LocalInvocationID.x % 64; | |
| const uint il = tid / 4; | |
| const uint ir = tid % 4; | |
| const uint ib = 4*i + ir; | |
| if (ib >= p.nel / QUANT_K_Q2_0) { | |
| return; | |
| } | |
| const uint b_idx = 256*i + QUANT_K_Q2_0*ir + 4*il; | |
| const uint bits = uint(data_a[ib].qs[il]); | |
| const float d = float(data_a[ib].d); | |
| data_b[b_idx ] = D_TYPE(d * (float(bits & 3u) - 1.0f)); | |
| data_b[b_idx + 1] = D_TYPE(d * (float((bits >> 2u) & 3u) - 1.0f)); | |
| data_b[b_idx + 2] = D_TYPE(d * (float((bits >> 4u) & 3u) - 1.0f)); | |
| data_b[b_idx + 3] = D_TYPE(d * (float(bits >> 6u) - 1.0f)); | |
| } | |