Skip to content
Computer vision

AGP-Qwen3.6-35B-A3B

robot control model

1 min readupdated

This specialized fine-tune of the Qwen/Qwen3.6-35B-A3B base model was trained on successful robot task trajectories. It generates agent responses and tool calls inside the AGP robot interface, using visual observations and feedback to plan actions. Built for robotics research and autonomous agent projects that require multimodal data processing.

Key features

  • Built on the Qwen3.6-35B-A3B architecture with 35.11B parameters.
  • Trained on 5,658 successful trajectories from the LIBERO, RoboTwin, VLA-Arena, and RoboCasa simulators.
  • Keeps the vision encoder frozen during training.
  • Supports an input context length of up to 65,536 tokens.

What you can do

  • Run the model via vLLM to test it on robotics tasks.
  • Benchmark control performance in the VLA-Arena or RoboTwin simulators.
  • Integrate it with custom robot interfaces using the provided system prompts.

Example

vllm serve Agent-as-Policy/AGP-Qwen3.6-35B-A3B \
  --served-model-name agp-qwen36-sft \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder \
  --reasoning-parser qwen3
Read moreDetails
Benchmarks: 57.5% success rate on VLA-Arena and 62.3% on RoboTwin.
Weights format: BF16 Safetensors (12 shards).
License: Apache 2.0.
Discuss Telegram 0 13

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 65 GB · ready files72 GB
  • FP8 / INT8recommendedweights 33 GB36 GB
  • 4-bitweights 18 GB21 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.

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.

    # Agent-as-Policy/AGP-Qwen3.6-35B-A3B
    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 Agent-as-Policy/AGP-Qwen3.6-35B-A3B --local-dir ~/models/AGP-Qwen3.6-35B-A3B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to AGP-Qwen3.6-35B-A3B
  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

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

Agent-as-Policy/AGP-Qwen3.6-35B-A3B
Author
Agent-as-Policy
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

130

Ask AI about this model

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

Comments

Computer vision

Script2Play-4B

Script2Play-4B