admin

admin synced commits to refs/heads/feat/smolvla-on-steerable at admin/lerobot from mirror 2026-05-21 11:39:50 +00:00

e98b6f726b feat(train): debug dump runs inference too, with parity check

f7747d02a9 feat(train): periodic LM-head prediction dump for live debugging

86ecd4bc2e add subtask memory training recipe

28b86449a2 fix(pi05): cast attention masks to model dtype

Compare 4 commits »

admin synced commits to refs/heads/main at admin/lerobot from mirror 2026-05-21 11:39:50 +00:00

f4b834844e Feat/clean can bus (#3526)

admin synced commits to refs/heads/feat/add-topreward-rm at admin/lerobot from mirror 2026-05-20 19:19:57 +00:00

f6ecb7b955 refactor(rewards): clean up TOPReward processor/model

admin synced commits to refs/heads/feat/clean-can-bus at admin/lerobot from mirror 2026-05-20 19:19:56 +00:00

6aa513f58e update exception catch

ec6c7f95c4 bugfix + %-style

Compare 2 commits »

admin synced commits to refs/heads/rebot-mit-mode at admin/lerobot from mirror 2026-05-20 19:19:56 +00:00

admin synced new reference refs/heads/rebot-mit-mode to admin/lerobot from mirror 2026-05-20 19:19:56 +00:00

admin synced commits to refs/heads/feat/smolvla-on-steerable at admin/lerobot from mirror 2026-05-20 19:19:56 +00:00

5bb2da4da6 fix(pi052): VQA target format = "label <loc><loc>" not "<loc><loc> label"

f7b989ad97 fix(pi052): read backbone dtype from q_proj, not first parameter

3b4376aa33 fix(pi052): cast attention bias to model dtype for bf16 inference

Compare 3 commits »

admin synced new reference refs/heads/docs/model-card-improvements to admin/lerobot from mirror 2026-05-20 11:10:02 +00:00

admin synced commits to refs/heads/feat/smolvla-on-steerable at admin/lerobot from mirror 2026-05-20 11:10:02 +00:00

34269a5d78 fix(pi052): register PaliGemma <loc> tokens so they tokenize as single ids

admin synced commits to refs/heads/auto/update-uv-lock at admin/lerobot from mirror 2026-05-20 11:10:01 +00:00

8ad6193549 chore(dependencies): update uv.lock

dfdc48a7f1 fix(datasets): bound VideoDecoderCache to prevent OOM on large datasets (#3614)

6a8878a639 fix(datasets): normalize shape=(1,) numeric values before HF encoding (#3344)

d38eb89f71 feat(video re-encoding): Adding utility and dataset edition tool for video re-encoding (#3611)

7ab4936b1b Add extensive language support (#3467)

Compare 6 commits »

admin synced commits to refs/heads/feat/language-annotation-pipeline at admin/lerobot from mirror 2026-05-20 11:10:01 +00:00

a0233f53f4 feat(annotate): default VLM to Qwen3.6-35B-A3B-FP8

admin synced commits to refs/heads/docs/model-card-improvements at admin/lerobot from mirror 2026-05-20 11:10:01 +00:00

admin synced commits to refs/heads/feat/smolvla-on-steerable at admin/lerobot from mirror 2026-05-20 02:59:50 +00:00

75507491bf fix(pi052): VQA <loc> conversion treats coords as 0-1000 normalized

88519cb14c fix(pi052): quantile-normalize actions before FAST tokenizer fit

bc0c993b25 fix(pi052): FAST tokenizer fit read actions from column, not ds[i]

ddf4bc2063 fix(pi052): knowledge insulation crashed on wrong _gated_residual import

b7317b6c29 test(pi052): round-trip coverage for VQA <loc> conversion

Compare 5 commits »

admin synced commits to refs/heads/feat/smolvla-on-steerable at admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

c026aed8f8 feat(pi052): train VQA spatial answers in PaliGemma <loc> format

e425dfd624 fix(processor): fallback to task message when recipe misses

15f79b5e5e fix(pi052): supervise an EOS token at the end of each text target

725ac95b0d feat(runtime): make the interactive runtime drive PI052 too

7b64e5498d revert(annotate): move memory + speech prompts to base PR (#3471)

Compare 6 commits »

admin synced commits to refs/heads/feat/add-topreward-rm at admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

admin synced new reference refs/heads/feat/add-topreward-rm to admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

admin synced commits to refs/heads/feat/language-annotation-pipeline at admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

2ea0da2d9f fix(annotate): tag uploaded dataset revision

134a707c7a feat(annotate): first-person memory narrative + shorter speech prompts

ce47075d6b feat(annotate): deterministic plan, single-frame VQA, dataset tagging

26013da699 feat(annotations): enforce imperative verb-first subtask phrasing

Compare 4 commits »

admin synced and deleted reference refs/heads/feat/language-columns at admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

admin synced and deleted reference refs/heads/feat/video-re-encode at admin/lerobot from mirror 2026-05-19 18:49:54 +00:00

admin synced commits to refs/heads/main at admin/lerobot from mirror 2026-05-19 18:49:53 +00:00

dfdc48a7f1 fix(datasets): bound VideoDecoderCache to prevent OOM on large datasets (#3614)

6a8878a639 fix(datasets): normalize shape=(1,) numeric values before HF encoding (#3344)

d38eb89f71 feat(video re-encoding): Adding utility and dataset edition tool for video re-encoding (#3611)

7ab4936b1b Add extensive language support (#3467)

ca8c60a0ed Set OpenCV fourcc after size and fps (#3620)

Compare 5 commits »