Instructions to use Ame233/lingbot-va-agilex with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Ame233/lingbot-va-agilex with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Ame233/lingbot-va-agilex", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
LingBot-VA AgileX SFT checkpoints
This repository contains task-specific Transformer checkpoints and one shared copy of the frozen VAE, UMT5 text encoder, and tokenizer.
Shared components
vae/, text_encoder/, and tokenizer/ are copied byte-for-byte from robbyant/lingbot-va-base, revision 68b7bc1b35da6ddc67ea94c4ceb58d768fbb3f9c. These components were unchanged during the SFT runs and are shared by all checkpoints in this repository.
The original weights are from the Robbyant LingBot-VA project, released under Apache License 2.0. See LICENSE, the upstream model card, and upstream code for attribution and model details. Per-file SHA256 hashes are recorded in shared_components.json.
Transformer checkpoints
| Task | Transformer directories |
|---|---|
| AgileX pour | checkpoint-800/transformer/, checkpoint-900/transformer/, checkpoint-1000/transformer/ |
| AgileX opendraw | agilex_opendraw/checkpoint-1000/transformer/ |
| AgileX pick | agilex_pick/checkpoint-1000/transformer/ |
Load the three shared components from the repository root and select the Transformer directory for the desired task and training step. Each shared directory retains its original file names and configuration; the text encoder includes all three weight shards and their index.
For example, the tokenizer can be loaded with T5TokenizerFast.from_pretrained("Ame233/lingbot-va-agilex", subfolder="tokenizer"). The VAE and text encoder similarly use subfolder="vae" and subfolder="text_encoder" with AutoencoderKLWan and UMT5EncoderModel.
Transformer loading requires the LingBot-VA implementation. Task-specific action normalization and execution configuration must be supplied separately; they are not replaced by these shared components.
- Downloads last month
- -