Instructions to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"

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 avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw"

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 "avlp12/Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Kimi-K2.7-Code-Alis-MLX-Dynamic-3.6bpw / figures /field.svg

avlp12

Kimi-K2.7-Code 3.6bpw MLX: weights + model card + figures

e9009a3 verified about 2 months ago

preview code

Raw

History Blame Contribute Delete

2.54 kB

	<svg viewBox="0 0 1100 600" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system, Helvetica, Arial, sans-serif">
	<rect width="1100" height="600" fill="#ffffff"/>
	<text x="50" y="56" font-size="34" font-weight="700" fill="#0f172a">Kimi-K2.7-Code MLX builds — size vs a 512 GB box</text>
	<text x="50" y="90" font-size="20" fill="#475569">Only sub-4-bit-expert builds fit a single 512 GB M3 Ultra</text>

	<!-- scale: x 300..1040 = 740px = 800 GB -> 1 GB = 0.925px ; origin x0=300 = 0 GB -->
	<!-- 512 GB line -->
	<line x1="773" y1="130" x2="773" y2="500" stroke="#16a34a" stroke-width="2.5" stroke-dasharray="6 5"/>
	<text x="773" y="124" font-size="17" font-weight="600" fill="#16a34a" text-anchor="middle">512 GB</text>
	<!-- 768 GB line -->
	<line x1="1010" y1="130" x2="1010" y2="500" stroke="#dc2626" stroke-width="2" stroke-dasharray="6 5"/>
	<text x="1010" y="124" font-size="17" font-weight="600" fill="#dc2626" text-anchor="middle">768 GB</text>

	<g font-size="20">
	<!-- pipenetwork 4bit-hiprec ~600 GB -->
	<text x="50" y="196" font-weight="600" fill="#0f172a">pipenetwork 4bit-hiprec</text>
	<text x="50" y="220" font-size="15" fill="#64748b">~5.0 bpw</text>
	<rect x="300" y="176" width="555" height="44" rx="6" fill="#94a3b8"/>
	<text x="867" y="204" font-size="18" fill="#dc2626">~600 GB ✗</text>

	<!-- spicyneuron 3.6bit ~460 GB -->
	<text x="50" y="296" font-weight="600" fill="#0f172a">spicyneuron 3.6bit</text>
	<text x="50" y="320" font-size="15" fill="#64748b">~3.6 bpw</text>
	<rect x="300" y="276" width="426" height="44" rx="6" fill="#60a5fa"/>
	<text x="738" y="304" font-size="18" fill="#475569">~460 GB · borderline</text>

	<!-- this build 465 GB -->
	<text x="50" y="396" font-weight="700" fill="#0f172a">this build (3.6bpw)</text>
	<text x="50" y="420" font-size="15" fill="#64748b">3.62 bpw · #907-fixed</text>
	<rect x="300" y="376" width="430" height="44" rx="6" fill="#4f46e5"/>
	<text x="742" y="404" font-size="18" font-weight="600" fill="#16a34a">465 GB ✓ · split 2× ✓✓</text>
	</g>

	<line x1="50" y1="470" x2="1050" y2="470" stroke="#e2e8f0" stroke-width="1.5"/>
	<text x="50" y="506" font-size="18" fill="#334155">The only build with experts re-quantized below the INT4 master and a verified two-machine pipeline path.</text>
	<text x="50" y="540" font-size="16" fill="#64748b">Sizes from each build's published config / card. The 600 GB build cannot hold a 512 GB box's worth of weights + cache.</text>
	</svg>