Skip to content
Language models

Wayfinder-qwen-eff

An efficient code search agent based on Qwen

2 min read

The model is built on Qwen3.6-35B-A3B (35 billion parameters, 3 billion active per token) and trained to solve code search tasks within codebases. After supervised fine-tuning, the authors performed 10 steps of reinforcement learning, rewarding the model for finding the correct files with minimal overhead. The result is an agent that uses roughly half as many tokens and search steps as the base version while maintaining the exact same recall. This model is useful for developers and engineers automating work with large repositories and looking to reduce LLM query costs.

Key features

  • Maintains a Recall@12 of 0.743 on complex long queries while cutting token usage in half.
  • Reduces the number of search steps by an average of 2.6 per session.
  • Trained using a combined reward for accuracy and efficiency.
  • Weights are provided in bf16 safetensors format with the standard tokenizer and chat template of the base model.

What you can do

  • Apply the agent to automated file search tasks and navigation across large codebases.
  • Set up efficient handling for long text prompts and technical descriptions of 120 words or more.
  • Integrate the model into development pipelines to reduce token costs during information retrieval.

Example

The model is used as a search sub-agent for working with a codebase. For best results, providing detailed, long-form task briefs is recommended, as performance and accuracy drop on short single-line prompts.

Read moreDetails
• Metrics: Recall@12 is 0.743, averaging 5.3 steps and 106k tokens per search compared to 7.9 steps and 213k for the base version.
• Training: LoRA (rank 64) in the first stage and 10 RL steps on a pool of 1,046 tasks.
• Limitations: efficiency drops on short single-line prompts; long descriptions are required.
• License: derivative model under the Muse Model License with mandatory attribution.
Discuss Telegram 0

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 67 GB · ready files68 GB
  • FP8 / INT8recommendedweights 33 GB35 GB
  • 4-bitweights 19 GB20 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.

Language models: with a 8192-token context.

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

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent RTX A6000

    $0.43 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # hweller1/wayfinder-qwen-35b-a3b-eff — RTX A6000, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve hweller1/wayfinder-qwen-35b-a3b-eff --max-model-len 8192 --quantization fp8
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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 Qwen3.6-35B-A3B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 4All on Hugging Face

About the model

hweller1/wayfinder-qwen-35b-a3b-eff
Author
hweller1
Category
Language models
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

Computer vision

GameDojo-4B

GameDojo-4B