cookbook(action/droid): align DROID recipe with GB200 reference reproduction#261
Merged
Merged
Conversation
Add a 'Verifying numerical reproducibility' section to the DROID/LIBERO action-policy finetune cookbook, pointing at cosmos-framework's deterministic action_policy_regression_test with the reference H200 LIBERO-10 loss series. Companion to cosmos-framework PR NVIDIA#86. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…8192, lr 2e-4) Pin the DROID repro TOML to the actual reference run shape instead of the single-node placeholder: data_parallel_replicate_degree=32 (HSDP 32x8 = 256 ranks), lr=2e-4, max_samples_per_batch=32, grad_accum_iter=1 -> global batch 8192, matching the LIBERO recipe's convention of encoding the true reference shape. Document the shape in the README/launcher and give the single-node scale-down override. Drops the earlier numerical-reproducibility section. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The action_policy_droid_nano experiment trains with format_prompt_as_json=True; call this out in the DROID recipe description and flag that eval must use the matching JSON prompt format. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keep the HSDP 32x8 reference described as the GB200 shape (64 nodes x 4) without recommending an H100/H200 equivalent layout. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fwd4
changed the title
…le, dataloader) Sync the DROID repro TOML with the framework copy: fold in loss_scale=10, scheduler.cycle_lengths=[100000], compile_tokenizer, and dataloader worker settings so the recipe reproduces the reference run without launcher overrides. Update the README description (cosine cycle 100000, generator loss_scale 10). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ychao-nvidia
reviewed
Jul 6, 2026
Collaborator
|
LGTM! |
ychao-nvidia
approved these changes
Jul 6, 2026
…erride Reformat the DROID single-node smoke-test EXTRA_TAIL_OVERRIDES onto multiple lines for readability, per @ychao-nvidia's suggestion on PR NVIDIA#261. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ychao-nvidia
approved these changes
Jul 6, 2026
fwd4
added a commit
to NVIDIA/cosmos-framework
that referenced
this pull request
Jul 7, 2026
…o recipe + numerical regression test (#86) ## Summary Adds **DROID action-policy post-training** on Cosmos3-Nano and a **numerical regression test** for the action-policy launches — the reference reproduction recipe for the DROID policy result. ## What's included - **Lazy DROID LeRobot dataset** — `cosmos_framework/data/generator/action/datasets/cosmos3_action_lerobot.py` (streaming `BaseActionLeRobotDataset`) + a rewritten `droid_lerobot_dataset.py` on top of it, plus `droid_lerobot_dataset_config.py`. Keys the versioned merged root; `use_success_only` resolves the `success/` split; eager `_register_sources()`. - **DROID Nano recipe** — `configs/base/experiment/action/posttrain_config/action_policy_droid_nano.py` + `examples/toml/sft_config/action_policy_droid_repro.toml`: res480, `joint_pos` 8D + `use_state`, JSON action prompt (`format_prompt_as_json=True`), CPU-side color jitter. The TOML pins the GB200 reference shape — HSDP 32×8 (256 ranks), global batch 8192, lr 2e-4, 10000 iters — and trains the generation + action heads from the public Cosmos3-Nano base. - **net_ema warm-start** — `checkpoint/dcp.py`: seed `net_ema` from `net` when `net_ema` is skipped on load (fresh action heads). - **Action-policy numerical regression test** — `tests/action_policy_regression_test.py`: the action-policy analogue of `tests/launch_regression_test.py`. Deterministic 10-iter re-run of the LIBERO + DROID launches (single-node, `--deterministic`, seed 42), asserting per-arch rank-0 loss goldens with a tolerance. LIBERO golden captured on the H200 CI arch; the DROID spec skips unless its (large, out-of-CI) dataset is supplied via `DROID_ROOT`. ## Reproduction The recipe reproduces the DROID action-policy result from the public Cosmos3-Nano base. The exact training code was validated by a from-scratch 64-GPU (GB200) run to 10k iterations, and the recipe TOML now encodes that run's configuration directly. ## Companion Cookbook PR: NVIDIA/cosmos#261 — the runnable DROID/LIBERO finetune cookbook that drives this recipe. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: lfengad <liangf@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Aligns the DROID action-policy finetune cookbook (
cookbooks/cosmos3/generator/action/finetune/) with the actual Cosmos3-Nano-Policy-DROID GB200 reference reproduction run.The DROID recipe files already exist on
main; this PR pins the recipe's run-level scalars to the real reference shape instead of the single-node placeholder, mirroring how the LIBERO recipe already encodes its reference config.Changes
action_policy_droid_repro.toml— encode the GB200 reference config:data_parallel_replicate_degree=32/data_parallel_shard_degree=8→ HSDP 32×8 = 256 ranks (GB200: 64 nodes × 4)lr=2.0e-04,max_samples_per_batch=32,grad_accum_iter=1→ global batch 8192max_iter=10000,cycle_lengths=[10000]format_prompt_as_json=True); note that eval must use the matching JSON prompt. Add the single-node scale-down override (replicate=1+grad_accum_iter=32keeps GBS 8192).launch_sft_action_policy_droid.sh— fix the stale "8× H100" / "8-GPU FSDP by default" comments to reflect the 256-rank reference shape.The TOML is kept byte-identical to the cosmos-framework copy (
examples/toml/sft_config/action_policy_droid_repro.toml).Notes
🤖 Generated with Claude Code