Downloads · 30 days
0
bi-lawm-bidir-stage2
GT-111/bi-lawm-bidir-stage2
bi-lawm-bidir-stage2 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.
Access
Public
Updated Aug 31, 2026
Repo size
144 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt76.8 GB · 53%
At a glance
- Task
- Robotics
- Access
- Public
- Created
- Aug 31, 2026
- Updated
- Aug 31, 2026
- SHA
- eac420be
Model card
From the Hugging Face model README
Bi-LaWM Bidir Stage2 — AgiBot + Intern-A1 Mixture
Repository: GT-111/bi-lawm-bidir-stage2
Intermediate checkpoints from the bidirectional Stage2 training run of Bi-LaWM on an AgiBot World + Intern-A1 mixture (200k-step plan).
Lineage
- Latent-action model initialized from the Stage1 38k BiLAM export (
export/step-038000/, consolidated single-filepytorch_model.pt). - Training resumed from the Stage1 lineage at step 15,000 and continues to the 200k-step plan.
- Milestones are published every 10k steps plus a snapshot of the latest checkpoint.
Checkpoints
| Hub path | Training step | Format |
|---|---|---|
checkpoints/step-010000/ | 10,000 | Full sharded FSDP training checkpoint (Stage1 phase of the lineage) |
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 |
checkpoints/step-055000/ | 55,000 | Latest snapshot, full sharded FSDP training checkpoint |
export/step-038000/ | 38,000 (Stage1) | Consolidated single-file pytorch_model.pt latent-action-model export |
Each training checkpoint contains the model shards, per-rank optimizer and RNG states, scheduler state, and trainer_state.json. These are resumable distributed-training checkpoints rather than merged inference-only weights.
Experiment
- Project: Bi-LaWM
- Datasets: AgiBot World + Intern-A1 mixture
- Stage: Stage2 bidirectional policy + latent-action training
- Training plan: 200k steps (discontinued)
- Final: 55k steps — training was stopped at 55k; the 50k milestone was exported as init for a downstream RoboTwin finetune
Loading
Use these checkpoints with the matching Bi-LaWM code and FSDP configuration. For inference or evaluation, export/merge the selected distributed checkpoint with the project checkpoint tooling first.
- Task
- Robotics
- Created
- Aug 31, 2026
- Updated
- Aug 31, 2026