Qwen3.6-27B-rust-v1-4bit.mlx

Qwen3.6-27B-rust-v1 4-bit MLX

A Rust-focused Qwen3.6-27B model for Apple Silicon, packaged in MLX.

Use it as a coding assistant for Rust projects: generating focused patches, explaining diffs, tightening tests, reading command output, and making small repo-aware edits. It is tuned for tool-calling workflows where the assistant has to inspect files, run commands, and keep changes narrow.

This is the plain 4-bit OptIQ variant. This plain package does not include native MTP tensors. It is the safer choice for general MLX loaders and clients that do not yet recognize the Qwen3.6 dense MTP layout.

Which Variant Should I Use?

• Use this repo if you want the plain 4-bit OptIQ compatibility package, or if your MLX runtime does not load native MTP weights.
• Use jedisct1/Qwen3.6-27B-rust-v1-8bit.mlx for the plain 8-bit sibling.
• Use jedisct1/Qwen3.6-27B-rust-v1-bf16.mlx for the plain BF16 full-precision sibling.
• Use jedisct1/Qwen3.6-27B-rust-v1-MTP-4bit.mlx if your runtime supports native MTP and you want the faster MTP path.

Usage

Requires mlx-lm:

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("jedisct1/Qwen3.6-27B-rust-v1-4bit.mlx")

messages = [
    {"role": "system", "content": "You are an expert Rust developer."},
    {"role": "user", "content": "Generate a focused patch that replaces unwrap() calls in parse_config() with proper error propagation."},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
print(response)

What It Is Good At

• Writing idiomatic Rust patches from a concise change request.
• Explaining Rust diffs in commit-message style.
• Following tool-calling workflows where it needs to inspect files before editing.
• Keeping changes focused instead of turning small fixes into broad rewrites.
• Working with tests, compiler errors, command output, and repository context.

Limitations

• Outputs should be reviewed before use, especially unsafe code, concurrency code, and changes that affect security boundaries.
• The model works best on focused Rust changes, tests, and explanations. Very large refactors may need to be split into smaller steps.
• Tool calling depends on the runtime and client preserving the chat template and tool schema format.