Skip to content
Computer vision

Qwen3.8-27B-pi-MLX-adapter

Coding LoRA for Apple Silicon

2 min read

This adapter (LoRA — a lightweight add-on built on top of a base model) ports the capabilities of the pi coding agent to the MLX architecture for running on Mac computers. Trained with AsyncGRPO, it handles terminal tasks and code generation efficiently. It lets you run a capable language model locally on MLX-supported hardware while saving memory by attaching the adapter to the base weights. The tool is aimed at developers and users who want offline, local coding workflows.

Key features

  • MLX optimization: Built specifically for the MLX framework on Apple Silicon devices.
  • Stronger coding performance: Delivers higher accuracy on terminal commands and programming tasks compared to the base model.
  • Token efficiency: Reduces the number of tool calls needed per task.
  • Flexible integration: Works on top of various quantized versions of the base model.

What you can do

  • Terminal automation: Generate complex bash scripts and system management commands.
  • Software development: Use the model as a local assistant for writing and debugging code.
  • Server deployment: Spin up a local API server for coding agents via mlx_lm.server.

Example

from huggingface_hub import snapshot_download
from mlx_lm import load, generate

adapter = snapshot_download("netprojects/Qwen3.8-27B-pi-MLX-adapter")
model, tokenizer = load("mlx-community/Qwen3.8-27B-4bit", adapter_path=adapter)
prompt = tokenizer.apply_chat_template([{"role": "user", "content": "Hello"}], add_generation_prompt=True,
                                       enable_thinking=False, tokenize=False)
print(generate(model, tokenizer, prompt=prompt, max_tokens=200))
Read moreDetails
• Benchmark results: Terminal-Bench 2.1 increased from 39.5% to 55.8%, SWE-bench Pro from 29.4% to 62.7%
• Adapter size: 638 MB in native float32 format
• Requirements: Requires mlx-lm version 0.32.2 or higher
• License: Apache-2.0
Discuss Telegram 0

Will it run?

Working out the memory needs

This is a LoRA: it only runs on top of the base model harness-hub/Qwen3.8-27B-pi-adapted. The base model's size is still being collected; the estimate will appear within a day.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # netprojects/Qwen3.8-27B-pi-MLX-adapter
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download harness-hub/Qwen3.8-27B-pi-adapted --local-dir ~/models/Qwen3.8-27B-pi-adapted
    hf download netprojects/Qwen3.8-27B-pi-MLX-adapter --local-dir ~/models/Qwen3.8-27B-pi-MLX-adapter
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.8-27B-pi-adapted
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

harness-hub/Qwen3.8-27B-pi-adapted

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. Qwen3.8-27Bfamily root
  2. Qwen3.8-27B-pi-adapted
  3. Qwen3.8-27B-pi-MLX-adapter

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

netprojects/Qwen3.8-27B-pi-MLX-adapter
Author
netprojects
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments