Skip to content
Robotics

xhand-crosshand-pi05

a pi0.5 fine-tune for robotics

2 min readupdated

This model provides a set of trained delta weights for the LeRobot-based pi0.5 robotics architecture, fine-tuned on a kettle manipulation task (Dexverse-GraspKettle-v0). It unifies three robotic hand types (sharpa, shadow, and wuji) into a shared 41-dimensional action space without relying on hand ID separation. During training, the vision module and action expert were updated while the PaliGemma language model remained frozen. It is useful for researchers and engineers working in robotics and robotic manipulator control.

Key features

  • Built on LeRobot's pi0.5 base model with a frozen language backbone
  • Supports cross-embodiment use across three different robotic hands (sharpa, shadow, wuji)
  • Distributed as delta weights (trainable.safetensors, 844.9M parameters) applied on top of the base model
  • Evaluated over 50 episodes per hand (150 runs total), achieving high success rates

What you can do

  • Apply the weights to fine-tune pi0.5-based robotics agents in simulators or real-world manipulation tasks
  • Merge the deltas with the lerobot/pi05_base model following the merge instructions to run the control policy
  • Reuse the included finger codec components and state-space configurations for your own robotic setups

Example

from huggingface_hub import snapshot_download
d = snapshot_download("Dexverse-CSI/xhand-crosshand-pi05", allow_patterns=["pooled3hand_nohot/Z/023670/*"])
base = snapshot_download("lerobot/pi05_base")
# Скрипты слияния и запуска доступны в репозитории проекта на GitHub (code/pi05_policy_server.py)
Read moreDetails
Benchmarks: 143/150 success rate for configuration Z and 135/150 for configuration D (50 episodes across each of the three hands)
647 tensors trained (844.9M parameters)
Requires the lerobot/pi05_base base model
License: unspecified (other); subject to the base model's license
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.8 GB · ready files2.5 GB
  • FP8 / INT8recommendedweights 0.9 GB1.5 GB
  • 4-bitweights 0.5 GB1.1 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 2060

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

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.

    # Dexverse-CSI/xhand-crosshand-pi05
    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 Dexverse-CSI/xhand-crosshand-pi05 --local-dir ~/models/xhand-crosshand-pi05
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to xhand-crosshand-pi05
  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

  1. paligemma-3b-pt-224family root
  2. pi05_base
  3. xhand-crosshand-pi05

New posts about versions of paligemma-3b-pt-224 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

Dexverse-CSI/xhand-crosshand-pi05
Author
Dexverse-CSI
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments