Commit History
-
LIBERO offline: TinyVLA v1 (2.5GB) ratio 0.092 vs pi0.5-libero-finetuned (13.8GB) ratio 0.057; GR00T not loadable
Verified signature603f504 -
Upload tinyvla_b200/renders_dagger_closedloop/cl_vla_stand.mp4 with huggingface_hub
Verified signaturead98bfb -
Upload tinyvla_b200/renders_dagger_closedloop/cl_vla_ground_pick.mp4 with huggingface_hub
Verified signatured14330e -
Upload tinyvla_b200/renders_dagger_closedloop/cl_vla_walking.mp4 with huggingface_hub
Verified signature0531549 -
Upload tinyvla_b200/renders_dagger_closedloop/cl_vla_sitstand.mp4 with huggingface_hub
Verified signature4102533 -
Upload tinyvla_b200/ckpt_dagger_iter0/README.md with huggingface_hub
Verified signaturecb22f7c -
DAgger iter0 checkpoint: first closed-loop STABLE behavior (sitstand 8s), loss 0.001-0.004
Verified signature50f53da -
Upload tinyvla_b200/renders_closedloop/cl_vla_stand.mp4 with huggingface_hub
Verified signature2e5baa2 -
Upload tinyvla_b200/renders_closedloop/cl_vla_walking.mp4 with huggingface_hub
Verified signatured696c89 -
Upload tinyvla_b200/renders_closedloop/cl_teacher_walking.mp4 with huggingface_hub
Verified signaturec5754fa -
VLA closed-loop driver (fixed head camera + auto-gain): falls <1s all behaviors — BC covariate shift on unstable system
Verified signature57af0e0 -
closed-loop MicroDuck harness: obs.rs 61D builder validated by ONNX teacher (stands 8s @10Hz replan)
Verified signature5a27506
-
duck FT eval (fixed expert): 13/21 better than zero-baseline
Verified signature8c01232 -
sim renders: GT vs duck-FT policy (FIXED expert), external camera
Verified signatured52a3c4 -
microduck fine-tune ckpt (8k steps on duck v2 data), expert keys fixed
Verified signature00c8545 -
fix expert keys (_orig_mod stripped)
Verified signaturef6de642 -
fix expert keys (_orig_mod stripped): prior file had a random flow head on load
Verified signature9ec62a4 -
CRITICAL fix: torch.compile'd expert saved as expert._orig_mod.* -> from_pretrained silently dropped it (random flow head at eval). Save the uncompiled module
Verified signaturef0fcec5 -
lekiwi zero-shot eval harness (hub streaming, 2x2 demo/descriptor grid)
Verified signaturea6a7adb -
zero-shot LeKiwi eval of final v4 ckpt: ratio 1.64 all conditions (demos and descriptor both inert)
Verified signaturec3a6904
-
FINAL v4-allchannels checkpoint @ 60k global steps (11.5M samples, 163-dataset physical-ai stream)
Verified signature13eb58f
-
microduck sim renders, external MuJoCo camera: GT teacher vs v4@20k side by side, offset-from-STAND action convention
Verified signature05e5dc3 -
v4-allchannels intermediate checkpoint @ ~20k global steps (3.8M samples over 163-dataset physical-ai stream) + README
Verified signature26c6a36 -
fix: HubEpisodeStream.set_epoch was missing — first worker-recycle epoch boundary crashed the run
Verified signature8ce1d5b
-
eval_microduck: zero-shot eval+render harness (v4 mode: demos from same behavior, duck descriptor)
Verified signature4dabbd2 -
train_fast: non-persistent hub workers (recycle each epoch to reset leaks)
Verified signatured990e88 -
streaming: episodes_per_epoch worker lifetime cap (dataloader workers leaked to OOM-kill on multi-hour runs twice)
Verified signature3859706 -
microduck zero-shot renders + metrics (v4 ckpt step_10000): all combos worse than zero-baseline, honest negative
Verified signaturec542a3d
-
wds mixture: unitree+nav converters (gnm/vamos), wds source branch in train_fast, AMP autoselect (V100 fp16), embedding expand 16->32, push/fetch/validate tooling, DATA.md
Verified signatureb152c62 -
v4 config: shrink shuffle buffer/workers — 5 images per sample (2 cams + K=3 demos) made 2048-buffer x12 workers OOM the dataloader
Verified signature75e042f -
specs: morph_key per dataset (6 robot families)
Verified signature411e21d -
GR00T family robot text prompts
Verified signature40b336d -
GR00T family morphology descriptors (estimates, flagged)
Verified signature52aa9f4 -
config: v4 all-channels on full physical-ai pool (smoke-verified, running)
Verified signatureb726c0a -
train_fast: v4 all-channels via hub stream (support passthrough, morph_text tokenization, guarded shard morph block)
Verified signatured9b76b2 -
streaming: in-context demo support cache (per-dataset episode ring, LRU), morphology+robot-text per sample
Verified signaturee23d6d9 -
B200.md: measured cost-per-sample table across GPUs (3090/4090/H100/B200), consumer-card build pitfalls
Verified signature69874c4 -
guard package init: keep bench/data paths importable without lerobot
Verified signature44d21d2
-
B200.md: measured smoke results (1xH100 495 s/s full script, 2xH100 DDP functional but interconnect-bound 340 s/s), spawn requirement, honest 8x caveat
Verified signature8d1c29a -
fix DDP hang: spawn dataloader workers (fork-after-CUDA/NCCL starved torchrun heartbeat), picklable Collate, fewer workers per rank
Verified signature4ac1f5f -
streaming: stats fallback via episode-parquet sampling for repos without meta stats (Tuned-Tasks)
Verified signaturee8f907d -
config: same yaml float fix
Verified signature754e7ad -
config: peak_flops in yaml-valid scientific notation
Verified signatureb905a69 -
fix: yaml 1.1 parses '2250.0e12' as string -> float() cast in MFU calc (crashed first log step)
Verified signaturefd625f9 -
B200.md: install order with all caught pitfalls (hf CLI patterns, py3.12, torchvision pair, causal-conv1d isolation)
Verified signature63e4224 -
requirements: py3.12, torch/tv pair, lerobot from git tag, triton>=3.7.1 (fla hopper bwd bug)
Verified signaturec4107f0 -
add data __init__
Verified signature162b571 -
add modules __init__
Verified signature5d5d9df -
add modules/embodiment.py
Verified signature0b12d75 -
add package __init__
Verified signatureb496d8a