Skip to content
Language models

Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2

optimized code completion model for AMD Ryzen AI processors

2 min readupdated

This is a custom quantized build of Qwen/Qwen2.5-Coder-3B (3B parameters) tailored for local fill-in-the-middle (FIM) code completion on AMD XDNA 2 NPU hardware accelerators via the FastFlowLM (FLM) framework. Standard uncalibrated 4-bit quantization degrades output quality too heavily on this hardware, so the author applied AWQ (activation-aware weight quantization) using the Pile dataset before repacking the weights to match the engine's requirements. It is built for developers and laptop owners with Ryzen AI 300 series or newer processors who want to run code autocomplete directly on their integrated NPU.

Key features

  • Base model: built on the popular Qwen2.5-Coder-3B language model tuned for coding tasks.
  • AWQ calibration: calibrated on the Pile corpus prior to Q4_0 quantization to preserve generation accuracy.
  • NPU support: packaged specifically to run on integrated AMD XDNA 2 neural processing units.
  • FIM integration: appends special fill-in-the-middle tokens at the end of the sequence and uses a pass-through chat template.

What you can do

  • Local code autocomplete: host a local code completion server on your AMD NPU without taxing a discrete GPU.
  • IDE integration: point your code editor to the local endpoint for fast, on-the-fly line and function suggestions.
  • Performance benchmarking: test code generation speed and quality directly on Ryzen AI accelerator hardware.

Example

Start the local model server via the flm-add utility and send a test FIM request:

uv tool install flm-add
flm-add nickto/Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2 \
  --tag qwen2.5-coder:3b-awq-pile-q4_0 --xclbin-from Qwen2.5-3B-Instruct-NPU2
export FLM_CONFIG_PATH="$HOME/.config/flm/model_list.json" FLM_XCLBIN_PATH="$HOME/.config/flm"
flm serve qwen2.5-coder:3b-awq-pile-q4_0

curl -s http://localhost:52625/v1/completions -H 'Content-Type: application/json' \
  -d '{"model":"qwen2.5-coder:3b-awq-pile-q4_0","prompt":"<|fim_prefix|>def add(a, b):\n    <|fim_suffix|>\n\nprint(add(1, 2))<|fim_middle|>","max_tokens":32,"temperature":0}'
Read moreDetails
SAFIM pass@1 benchmark (control / block / api): this build scores 54.7 / 39.5 / 65.5, compared to 5.3 / 3.9 / 18.1 without AWQ on FLM.
Hardware: AMD Ryzen AI processor with XDNA 2 architecture (Ryzen AI 300 series or newer).
Environment: FastFlowLM v1.0.11.
License: Qwen RESEARCH LICENSE AGREEMENT (Alibaba Cloud, non-commercial use only).
Discuss Telegram 0 29

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

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.

    # nickto/Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2
    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 nickto/Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2 --local-dir ~/models/Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2
    # 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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

nickto/Qwen2.5-Coder-3B-AWQ-Pile-Q4_0-NPU2
Author
nickto
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

290

Ask AI about this model

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

Comments