Instructions to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-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 hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
# Run inference directly in the terminal:
llama cli -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
# Run inference directly in the terminal:
llama cli -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

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 hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
# Run inference directly in the terminal:
./llama-cli -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

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 hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

Use Docker

docker model run hf.co/hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

LM Studio
Jan

vLLM

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hellohazime/Kimi-K3-REAP640ja-IQ1_S-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": "hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

Ollama
How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with Ollama:
```
ollama run hf.co/hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
```

Unsloth Studio

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF to start chatting

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

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": "hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with Docker Model Runner:
```
docker model run hf.co/hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S
```

Lemonade

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

Run and chat with the model

lemonade run user.Kimi-K3-REAP640ja-IQ1_S-GGUF-IQ1_S

List all available models

lemonade list

Hermes Agent

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-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 hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

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 hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S

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 "hellohazime/Kimi-K3-REAP640ja-IQ1_S-GGUF:IQ1_S" \
  --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"

hellohazime commited on 11 days ago

Commit

6448ff8

verified ·

1 Parent(s): b95a50d

Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +12 -7

README.md CHANGED Viewed

@@ -73,15 +73,20 @@ Honest scorecard: exactly what has been measured, and what has not.
 | Pruning is lossless for surviving experts | identity-prune is byte-identical (pinned by tests); router/norms stay F32 |
 | Generation settings disclosed | max_tokens 4096, thinking_effort low, temp 1.0, top-p 0.95, identical for both builds; 4/100 answers hit the 4096 cap. A 16k-budget recheck rescued none of them: 2 re-ran to clean completion *within* the original budget (stochastic thinking runaways at temp 1.0), 1 was still empty at 16k (114k chars of thinking), 1 hit 16k again in the answer body. The cap is not the bottleneck |
-**Not verified:**
-| open question | status |
 |---|---|
-| Agentic use (Kimi Code CLI, tool calling) | **never tested on this build** — the SWE-Lancer results on the sibling's card do not transfer |
-| Coding benchmarks | not measured; perplexity says expect degradation |
-| Japanese factual accuracy | the judge flagged factual errors even in fluent answers; 1.6-bit experts are fluent before they are precise |
-| Long-context quality | ELYZA prompts are short; 131k context is configured but unexercised here |
-| Vision | mmproj not included; text tensors only |
 ## Download & run

 | Pruning is lossless for surviving experts | identity-prune is byte-identical (pinned by tests); router/norms stay F32 |
 | Generation settings disclosed | max_tokens 4096, thinking_effort low, temp 1.0, top-p 0.95, identical for both builds; 4/100 answers hit the 4096 cap. A 16k-budget recheck rescued none of them: 2 re-ran to clean completion *within* the original budget (stochastic thinking runaways at temp 1.0), 1 was still empty at 16k (114k chars of thinking), 1 hit 16k again in the answer body. The cap is not the bottleneck |
+**Verified operating envelope** — what this card's numbers actually cover:
+| measured at | detail |
 |---|---|
+| Single-turn Japanese generation | ELYZA-tasks-100, max_tokens 4096, thinking low, temp 1.0 / top-p 0.95 — the rubric/pairwise numbers above |
+| Perplexity | 2048-token windows, 4 domains (ja / en / code / zh table above) |
+| Serving | 131,072 context configured and stable; ELYZA prompts exercise only the short end of it |
+| Modality | text tensors only (no mmproj shipped) |
+Agentic tool-calling sessions have been run **only on the en+code sibling** —
+its SWE-Lancer numbers belong to that build alone. The code-perplexity
+doubling above is the honest predictor for coding work here, and the judge
+flagged factual slips inside fluent Japanese: 1.6-bit experts are fluent
+before they are precise.
 ## Download & run