Skip to content
Robotics

pi05-libero-goal-bowl_on_plate

teaching a robot to place a bowl on a plate

2 min readupdated

A robotic arm control model fine-tuned for a specific household task: picking up a bowl and placing it on a plate in the LIBERO-goal simulator. It builds on the 4.14B-parameter lerobot/pi05_base architecture using full SFT (Supervised Fine-Tuning — updating all model weights without freezing layers or using LoRA). Unlike the general-purpose base version, this checkpoint is tailored strictly to a single action and includes intermediate saves alongside optimizer states, not just the final weights. It is designed for robotics researchers and LeRobot framework users looking to reproduce experiments or resume training.

Key features

  • Full fine-tuning: all 4,143,404,816 model parameters were trained in FP32 precision with BF16 autocast and the AdamW optimizer.
  • Vision-action processing: takes 256×256 images from two cameras (plus one masked input) and generates 50-step action sequences using 10 flow-matching iterations (a method for generating smooth trajectories).
  • Complete checkpoint set: the repository includes saves at steps 250, 500, 750, and 1000, bundled with training_state so you can resume training right where it left off.
  • PaliGemma compatibility: handles text instructions using the official Big Vision PaliGemma vocabulary.

What you can do

  • Run the PI05Policy in the LIBERO-goal simulation environment (task ID: 8) to evaluate how reliably it places the bowl onto the plate.
  • Load any intermediate checkpoint along with its optimizer state to continue fine-tuning on your own demonstrations in LeRobot.
  • Compare task performance across different training stages using the raw evaluation logs included in the repository.

Example

from huggingface_hub import snapshot_download
from lerobot.policies.pi05.modeling_pi05 import PI05Policy

root = snapshot_download("qrvmil/pi05-libero-goal-bowl_on_plate-20261116", allow_patterns="checkpoints/000250/pretrained_model/*")
policy = PI05Policy.from_pretrained(root + "/checkpoints/000250/pretrained_model")
Read moreDetails
Success Rate (5 trials, 300-step limit): 20% (1/5) at step 250, 0% at step 500, and 60% (3/5) at step 750; step 1000 evaluation is currently in progress.
Training: 1000 steps across 49 demonstrations (4,563 frames), batch size 8, learning rate 5e-5 with cosine decay.
Control: relative end-effector control at 20 Hz, executing 10 out of 50 predicted actions per step.
Discuss Telegram 1 30

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 15 GB · ready files17 GB
  • BF16weights 7.7 GB9.0 GB
  • FP8 / INT8recommendedweights 3.9 GB4.7 GB
  • 4-bitweights 2.2 GB2.9 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 · F32 — from RTX 4090

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.

    # qrvmil/pi05-libero-goal-bowl_on_plate-20261116
    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 qrvmil/pi05-libero-goal-bowl_on_plate-20261116 --local-dir ~/models/pi05-libero-goal-bowl_on_plate-20261116
    # The model is in ~/models. How to run it is in the model card (link above).
  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. pi05-libero-goal-bowl_on_plate-20261116

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

qrvmil/pi05-libero-goal-bowl_on_plate-20261116
Author
qrvmil
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

301

Ask AI about this model

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

Comments