Skip to content
Computer vision

Qwen3.8-27B-opencode-adapted-GGUF

a code model quantized for llama.cpp

2 min readupdated

This model is a quantized GGUF version of the fine-tuned Qwen/Qwen3.8-27B model, optimized for coding and terminal tasks. It was created by merging the base model with a LoRA adapter trained using reinforcement learning in the Harbor and OpenEnv environments. The model supports built-in Multi-Token Prediction (MTP) to speed up generation during local execution. It is useful for developers and engineers running a local AI coding assistant.

Key features

  • Ready-to-use GGUF format: Models are quantized to run via llama.cpp without needing separate LoRA attachments.
  • Multi-Token Prediction (MTP): Built-in multi-token prediction heads accelerate generation without a separate draft model.
  • Various quantization options: Versions ranging from compact Q4 to maximum Q8 are available.
  • Code specialization: Trained for engineering tasks and terminal workflows.

What you can do

  • Local execution: Deploy the model in applications and utilities supporting llama.cpp and the GGUF format.
  • Software development: Use it for code generation, debugging, and automating terminal tasks.
  • Multimodal tasks: Attach a separate projection file mmproj-bf16.gguf to work with images.

Example

Download the main weights file for your desired quantization (for example, Q4_K_M) alongside the mmproj-bf16.gguf projection file, then run them through a compatible llama.cpp engine for coding work.

Read moreDetails
Adapter benchmarks: SWE-bench Pro (hard 51) — 51.0% (compared to 35.3% for base), Terminal-Bench 2.1 — 52.4% (compared to 46.4%).
File sizes: From 15.8 GB to 29 GB for the main weights and 931 MB for mmproj.
Features: Contains built-in MTP tensors to speed up inference.
Discuss Telegram 3 266

Will it run?

Runs in GGUF Q5_K_M

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

  • GGUF Q8_0recommendedweights 27 GB · ready files30 GB
  • GGUF Q5_K_Mweights 18 GB · ready files21 GB
  • GGUF Q5_K_Sweights 18 GB · ready files20 GB
  • GGUF Q4_K_Mweights 16 GB · ready files18 GB
  • GGUF Q4_K_Sweights 15 GB · ready files17 GB

Tip: go 8-bit.GGUF Q8_0 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q8_0 — from RTX 5090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

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.

    # prithivMLmods/Qwen3.8-27B-opencode-adapted-GGUF
    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 prithivMLmods/Qwen3.8-27B-opencode-adapted-GGUF --local-dir ~/models/Qwen3.8-27B-opencode-adapted-GGUF
    # The model is in ~/models. How to run it is in the model card (link above).
  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

  1. Qwen3.8-27Bfamily root
  2. Qwen3.8-27B-opencode-adapted
  3. Qwen3.8-27B-opencode-adapted-GGUF

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

prithivMLmods/Qwen3.8-27B-opencode-adapted-GGUF
Author
prithivMLmods
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

2663

Ask AI about this model

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

Comments