Skip to content
Computer vision

Qwen3.8-27B-pi-adapted

LoRA adapter for coding agents

2 min readupdated

This adapter (a lightweight add-on built on top of the base model) was trained using reinforcement learning to act as the "brain" for the pi coding agent on software engineering and terminal tasks. The model significantly boosts accuracy on complex engineering challenges, delivering results comparable to the base model's maximum reasoning mode while generating 4.5 times fewer tokens. This release will benefit developers and researchers using local LLMs for coding automation and terminal workflows.

Key features

  • High accuracy: performance gains on SWE-bench Pro and Terminal-Bench 2.1 compared to the base model.
  • Token efficiency: matches the accuracy level of the base model's heavy reasoning mode without enabling the special thinking mode.
  • Fewer steps: requires half as many interaction turns to solve similar tasks.
  • Engineering-ready: optimized specifically for agent tools and terminal calls.

What you can do

  • Run a coding agent: connect the model via vLLM or transformers serve to power the local pi agent.
  • Fix bugs: automate finding and fixing bugs across code repositories using terminal turns.
  • Test benchmarks: evaluate quality on SWE-bench Pro or Terminal-Bench 2.1.
  • Use GGUF: run a quantized version of the adapter locally using llama.cpp or Llama.app on macOS.

Example

vllm serve Qwen/Qwen3.8-27B \
    --enable-lora --max-lora-rank 32 \
    --lora-modules qwen3.8-pi=harness-hub/Qwen3.8-27B-pi-adapted \
    --language-model-only --enable-prefix-caching --mamba-cache-mode align \
    --enable-auto-tool-choice --tool-call-parser qwen3_xml --reasoning-parser qwen3 \
    --default-chat-template-kwargs '{"enable_thinking": false}' \
    --max-model-len 131072
Read moreDetails
• SWE-bench Pro (hard 51): accuracy jumped from 39.2% to 64.7%.
• Terminal-Bench 2.1: accuracy increased from 39.5% to 55.8%.
• Training: the adapter uses a LoRA rank of 32 and was trained on pi agent trajectories.
• Limitations: recommended to use with the reasoning mode disabled (`enable_thinking: false`).
Examples and videos
Discuss Telegram 0 18

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 52 GB · ready files58 GB
  • FP8 / INT8recommendedweights 26 GB30 GB
  • 4-bitweights 15 GB17 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

This is a LoRA: counted together with its base model Qwen3.8-27B.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # personaxis/Qwen3.8-27B-pi-adapted
    # ===== 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 Qwen/Qwen3.8-27B --local-dir ~/models/Qwen3.8-27B
    hf download personaxis/Qwen3.8-27B-pi-adapted --local-dir ~/models/Qwen3.8-27B-pi-adapted
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.8-27B
  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

Qwen/Qwen3.8-27B

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

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

personaxis/Qwen3.8-27B-pi-adapted
Author
personaxis
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments