Skip to content
Language models

sid1-qwen3-4b

a Qwen3-based search agent

2 min readupdated

The sid1-qwen3-4b model is a specialized fine-tune of the base Qwen/Qwen3-4B model (4 billion parameters), trained using Group Relative Policy Optimization (GRPO). Instead of answering questions directly, the model acts as a search agent: it retrieves relevant documents from a database using semantic and keyword search, performs multiple search steps, and returns a ranked list of document IDs. The development is based on the SID-1 concept and adapted for three dataset benchmarks.

Key features

  • Unsupervised reinforcement learning (GRPO) where the reward is the NDCG metric of the ranked document list.
  • Use of four tools: semantic search, keyword search (BM25), document reading, and final report.
  • Improved NDCG and recall scores compared to the base Qwen3-4B model.
  • High rate of correct tool calls (99.9%) and report completions.

What you can do

  • Integrate the model as a retrieval step in Question-Answering (QA) systems for complex multi-step queries.
  • Use the demo space on Hugging Face Spaces to test the agent and view recorded search runs.
  • Study the training loop code and reference agent for your own search agent experiments.

Example

import json, re, torch
from transformers import AutoTokenizer, AutoModelForCausalLM

repo = "masterofaudio2077/sid1-qwen3-4b"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, dtype=torch.bfloat16, device_map="auto").eval()

SYSTEM_PROMPT = (
    "You are a retrieval agent. You are given a question. Do not answer it. Find the documents in the collection that are needed to answer it "
    "and report their ids, most relevant first."
)

msgs = [{"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": "When did the luxury division of the maker of the Toyopet Master change the RX 350 body style?"}]
prompt = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=True, tokenize=False)
Read moreDetails
Benchmarks: NDCG 0.684 (1 sample), 0.729 (4-sample fusion)
Hardware requirements: peak VRAM consumption during training was 32.6 GiB
Limitations: early stopping at step 175 out of the planned 400, works only with English and a specific tool schema
License: Apache 2.0
Discuss Telegram 0 37

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 7.5 GB · ready files9.2 GB
  • FP8 / INT8recommendedweights 3.7 GB5.5 GB
  • 4-bitweights 2.1 GB3.8 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 3080

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

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 A30

    $0.26 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.

    # masterofaudio2077/sid1-qwen3-4b — A30, 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 masterofaudio2077/sid1-qwen3-4b --max-model-len 8192 --quantization fp8
    Add more LoRAs to sid1-qwen3-4b
  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

  1. Qwen3-4B-Basefamily root
  2. Qwen3-4B
  3. sid1-qwen3-4b

New posts about versions of Qwen3-4B-Base (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
Merges 1All on Hugging Face

About the model

masterofaudio2077/sid1-qwen3-4b
Author
masterofaudio2077
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

370

Ask AI about this model

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

Comments

Computer vision

Qwen3.81-27B

Qwen3.81-27B