Views
No views yet
{base}-{size}-ccmcp-{version}
ccmcp = Claude Chrome MCPadapters.safetensors - LoRA adapter weightsadapter_config.json - LoRA configurationsmollm-135m-ccmcp-v1-f16.gguf - GGUF F16 format for llama.cpp/Ollamacheckpoints/ - Training checkpoints1from mlx_lm import load, generate
2from mlx_lm.sample_utils import make_sampler
3
4model, tokenizer = load(
5 "mlx-community/SmolLM-135M-Instruct-4bit",
6 adapter_path="pierretokns/smollm-135m-ccmcp-v1"
7)
8
9messages = [
10 {"role": "system", "content": "You are a browser automation assistant with MCP tools."},
11 {"role": "user", "content": "Go to google.com"}
12]
13
14prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
15sampler = make_sampler(temp=0.1)
16
17response = generate(model, tokenizer, prompt=prompt, max_tokens=150, sampler=sampler)
18print(response)1# Download GGUF from this repo
2# Create Modelfile:
3cat > Modelfile << 'EOF'
4FROM ./smollm-135m-ccmcp-v1-f16.gguf
5PARAMETER num_ctx 8192
6PARAMETER temperature 0.1
7SYSTEM "You are a browser automation assistant with MCP tools."
8EOF
9
10# Create and run
11ollama create smollm-135m-ccmcp-v1 -f Modelfile
12ollama run smollm-135m-ccmcp-v1 "Go to google.com"1ANTHROPIC_BASE_URL=http://localhost:11434 \
2ANTHROPIC_AUTH_TOKEN=ollama \
3ANTHROPIC_API_KEY=ollama \
4claude --model smollm-135m-ccmcp-v1