Add files using upload-large-folder tool

d65ad3b verified 4 months ago

2.81 kB

license: apache-2.0
base_model:
  - Qwen/Qwen3.6-27B
library_name: mlx
tags:
  - mlx
  - qwen3.6
  - qwen3_5
  - apple-silicon
  - tool-calling
  - go
  - golang

Qwen3.6-27B-go-v1 MTP BF16 MLX

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

Use it as a coding assistant for Go 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 native-MTP BF16 full-precision variant. This native-MTP package includes the model's MTP weights for runtimes that support the Qwen3.6 dense MTP layout, such as oMLX. If a loader rejects MTP weights, use the matching plain repo instead.

Which Variant Should I Use?

Use this repo if your runtime supports native MTP for dense Qwen3.6 MLX models and you want the native-MTP BF16 full-precision package.
Use jedisct1/Qwen3.6-27B-go-v1-MTP-4bit.mlx for the native-MTP 4-bit OptIQ sibling.
Use jedisct1/Qwen3.6-27B-go-v1-MTP-8bit.mlx for the native-MTP 8-bit sibling.
Use jedisct1/Qwen3.6-27B-go-v1-bf16.mlx if your runtime does not support native MTP weights.

Usage

Requires mlx-lm:

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("jedisct1/Qwen3.6-27B-go-v1-MTP-bf16.mlx")

messages = [
    {"role": "system", "content": "You are an expert Go developer."},
    {"role": "user", "content": "Generate a focused patch that replaces panic() in loadConfig() with returned errors and table-driven tests."},
]
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 Go patches from a concise change request.
Explaining Go 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 concurrency code, unsafe/cgo code, and changes that affect security boundaries.
The model works best on focused Go 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.