Skip to content
Language models

qwen2.5-3b-toolcall-abstain-v6

accurate tool calling with abstention

2 min readupdated

This new language model, based on Qwen2.5-3B-Instruct, is fine-tuned for accurate tool calling and working with JSON schemas. Beyond generating valid function calls, it knows when to abstain if no suitable tool is available, or ask clarifying questions when required data is missing. Built for developers creating local AI agents and automation systems, the repository includes the merged model, a 4-bit version for Apple Silicon (via MLX), and a LoRA adapter.

Key features

  • Built for agents: optimized for complex function-selection scenarios and filtering out irrelevant distractor tools.
  • Reliable abstention: significantly outperforms the base model at recognizing when no tool call is needed.
  • Mac-ready: includes a 4-bit mlx/ build to run efficiently on Apple Silicon via the mlx-lm library.
  • Efficient training: trained using QLoRA on public datasets (xLAM, Glaive, Hermes, ToolACE).

What you can do

  • Build a local agent: connect the model to external APIs, databases, or tools through a custom pipeline.
  • Automate tasks: use the model for structured function calling in bots and services.
  • Run on Apple Silicon: deploy the 4-bit version locally on a Mac for fast inference.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

m = 'siva-0104/qwen2.5-3b-toolcall-abstain-v6'
tok = AutoTokenizer.from_pretrained(m)
model = AutoModelForCausalLM.from_pretrained(m, torch_dtype='auto')
Read moreDetails
Benchmarks: BFCL simple exact match — 89.7%, BFCL irrelevance abstained — 87.1%, xLAM exact — 77.3%.
Limitations: English only; when parameters are missing, the model occasionally guesses values instead of asking follow-up questions; retriever performance degrades when thousands of tools are available.
Training: QLoRA, 4-bit NF4, LoRA r=16 alpha=32, 1 epoch on a Kaggle T4 in 169 minutes.
License: inherits the Qwen Research License terms.
Discuss Telegram 0 315

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 11 GB · ready files12 GB
  • FP8 / INT8recommendedweights 8.6 GB9.5 GB
  • 4-bitweights 7.4 GB8.2 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 3080 · BF16 — from RTX 5080

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

This is a LoRA: counted together with its base model Qwen2.5-3B-Instruct.

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.

    # siva-0104/qwen2.5-3b-toolcall-abstain-v6 — A30, FP8 / INT8
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    # 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 Qwen/Qwen2.5-3B-Instruct --max-model-len 8192 --quantization fp8 --enable-lora --max-lora-rank 64 --lora-modules qwen2.5-3b-toolcall-abstain-v6=siva-0104/qwen2.5-3b-toolcall-abstain-v6
    Add more LoRAs to Qwen2.5-3B-Instruct
  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

Requires the base model

Qwen/Qwen2.5-3B-Instruct

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

  1. Qwen2.5-3Bfamily root
  2. Qwen2.5-3B-Instruct
  3. qwen2.5-3b-toolcall-abstain-v6

New posts about versions of Qwen2.5-3B (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 3All on Hugging Face

About the model

siva-0104/qwen2.5-3b-toolcall-abstain-v6
Author
siva-0104
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

3150

Ask AI about this model

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

Comments