Text Generation
GGUF
German
English
sakura
sakura-micro
llama.cpp
Mixture of Experts
mimo
expert-pruning
ssd-streaming
strix-halo
experimental
imatrix
conversational
Instructions to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF 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 webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF 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 webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS # Run inference directly in the terminal: llama cli -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS # Run inference directly in the terminal: llama cli -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS # Run inference directly in the terminal: ./llama-cli -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS # Run inference directly in the terminal: ./build/bin/llama-cli -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Use Docker
docker model run hf.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
- LM Studio
- Jan
- vLLM
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
- Ollama
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with Ollama:
ollama run hf.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
- Unsloth Desktop
- Pi
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with Docker Model Runner:
docker model run hf.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
- Lemonade
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Run and chat with the model
lemonade run user.Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF-IQ2_XS
List all available models
lemonade list
- Hermes Agent
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF:IQ2_XS" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
Download runtime/hot_lock_list.py from webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF: direct link, hf CLI and curl.
- Browser
- Download file 2.18 kB
-
https://huggingface.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF/resolve/main/runtime/hot_lock_list.py
- Command line
-
hf download hf://webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF/runtime/hot_lock_list.py
-
curl -L -o hot_lock_list.py https://huggingface.co/webmp3/Sakura-MiMo-V2.6-Flash-MOPD-P160-GGUF/resolve/main/runtime/hot_lock_list.py
2.18 kB
| #!/usr/bin/env python3 | |
| """Write an LLAMA_EXPERT_LOCK_LIST file: the most used kept experts of the streamed layers. | |
| Slots are positions in the pruned model's expert dimension = index in the ascending list of kept expert ids | |
| (the order prune_experts.py writes). Usage counts come from mimo_expert_importance.json (count_<source>). | |
| usage: python hot_lock_list.py --importance mimo_expert_importance.json --prune prune_..._keep144.json \ | |
| --layers 14-30 --frac 0.6 [--source onpol] --out hot60.txt | |
| """ | |
| import argparse | |
| import json | |
| def main(): | |
| ap = argparse.ArgumentParser() | |
| ap.add_argument("--importance", required=True) | |
| ap.add_argument("--prune", required=True, help="prune list of the model ({'pruned': {layer: [ids]}})") | |
| ap.add_argument("--layers", required=True, help="streamed layers, e.g. 14-30") | |
| ap.add_argument("--frac", type=float, default=0.6, help="share of kept experts to lock per layer") | |
| ap.add_argument("--source", default="onpol", help="usage source: onpol, de, baekpica, ...") | |
| ap.add_argument("--n-expert", type=int, default=256) | |
| ap.add_argument("--out", required=True) | |
| a = ap.parse_args() | |
| imp = json.load(open(a.importance, encoding="utf-8"))["layers"] | |
| pruned = json.load(open(a.prune, encoding="utf-8"))["pruned"] | |
| lo, hi = (int(x) for x in a.layers.split("-")) | |
| cover = [] | |
| with open(a.out, "w", encoding="ascii") as f: | |
| for L in range(lo, hi + 1): | |
| kept = sorted(set(range(a.n_expert)) - set(pruned[str(L)])) | |
| use = {r["expert"]: r.get(f"count_{a.source}", 0.0) for r in imp[str(L)]} | |
| order = sorted(kept, key=lambda e: -use.get(e, 0.0)) | |
| hot = order[: round(a.frac * len(kept))] | |
| slots = sorted(kept.index(e) for e in hot) | |
| f.write(f"{L} " + " ".join(map(str, slots)) + "\n") | |
| tot = sum(use.get(e, 0.0) for e in kept) or 1.0 | |
| cover.append(sum(use.get(e, 0.0) for e in hot) / tot) | |
| print(f"{a.out}: layers {lo}-{hi}, {a.frac:.0%} of kept experts per layer, " | |
| f"covers {sum(cover) / len(cover):.1%} of {a.source} routings (min {min(cover):.1%})") | |
| if __name__ == "__main__": | |
| main() | |