Downloads · 30 days
0
bi-lawm-stage2-only-ablation
GT-111/bi-lawm-stage2-only-ablation
bi-lawm-stage2-only-ablation is a robotics model from GT-111. Use it for the robotics task on the model card, and read the license before you ship it in a product.
Intermediate checkpoints from the Stage2-only unidirectional ablation of Bi-LaWM.
Access
Public
Updated Aug 31, 2026
Repo size
122 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt71 GB · 58%
At a glance
- Task
- Robotics
- Access
- Public
- Created
- Aug 27, 2026
- Updated
- Aug 31, 2026
- SHA
- 046f21c9
Model card
From the Hugging Face model README
Bi-LaWM Stage2-Only Ablation
Intermediate checkpoints from the Stage2-only unidirectional ablation of Bi-LaWM.
This repository contains the Stage2 policy-training trajectory at 10k-step intervals. The run was resumed across multiple launcher runs, but all checkpoints belong to the same training lineage. The 50k checkpoint was used to initialize the downstream RoboTwin fine-tuning experiment.
Checkpoints
| Hub path | Training step | Format |
|---|---|---|
checkpoints/step-010000/ | 10,000 | Full sharded FSDP training checkpoint |
checkpoints/step-020000/ | 20,000 | Full sharded FSDP training checkpoint |
checkpoints/step-030000/ | 30,000 | Full sharded FSDP training checkpoint |
checkpoints/step-040000/ | 40,000 | Full sharded FSDP training checkpoint |
checkpoints/step-050000/ | 50,000 | Full sharded FSDP training checkpoint |
export/step-050000/ | 50,000 | Consolidated single-file pytorch_model.pt export |
Each directory contains 27 files and includes trainer_state.json. These are resumable training checkpoints rather than merged inference-only weights.
Experiment
- Project: Bi-LaWM
- Variant: Stage2-only unidirectional ablation
- Training plan: 200k steps
- Published interval: every 10k steps through 50k
- Downstream use: the 50k checkpoint was exported (
export/step-050000/pytorch_model.pt) and used as the initialization for the RoboTwin Stage3 fine-tuning run (GT-111/bi-lawm-robotwin-stage3-finetune)
Notes
The checkpoint directories preserve their original FSDP layout. Load them with the matching Bi-LaWM training code and distributed configuration. For inference or downstream initialization, export/merge the selected checkpoint with the project's checkpoint export tooling first.
- Task
- Robotics
- Created
- Aug 27, 2026
- Updated
- Aug 31, 2026