Skip to content

Actions: physical-hardware-intelligence/phi

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
73 workflow runs
73 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
calibration: track the phi_leader arm ci #73: Commit 1b1127b pushed by BrutalCaeser
19s main
main
19s
act-why: the decoder's self-attention is provably inert (S10a) ci #72: Commit b818d74 pushed by BrutalCaeser
16s main
main
16s
docs/theory: ACT — why it is shaped like that ci #71: Commit 9472322 pushed by BrutalCaeser
16s main
main
16s
docs/theory: DP-transformer and SmolVLA teardowns, navigable by model ci #70: Commit 9e2a486 pushed by BrutalCaeser
21s main
main
21s
k-bucketed eval settles the DP-T A/B: paper is underfitting, not diff… ci #69: Commit 80619dd pushed by BrutalCaeser
15s main
main
15s
eval_by_k: fix import path caught by the smoke test ci #68: Commit 6dd5282 pushed by BrutalCaeser
20s main
main
20s
eval: k-bucketed held-out loss, + resume the paper arm to 60k ci #67: Commit aee1f1b pushed by BrutalCaeser
22s main
main
22s
DP-transformer A/B results: regularisation controls the overfit, and … ci #66: Commit 2995a03 pushed by BrutalCaeser
19s main
main
19s
calibration results: DP-transformer width is free, and data_s is not … ci #65: Commit 36d9524 pushed by BrutalCaeser
16s main
main
16s
hpc: DP-transformer regularisation A/B + pre-registration ci #64: Commit 25f2ada pushed by BrutalCaeser
18s main
main
18s
models: publish both Diffusion Policy runs -- the 20k checkpoint, not… ci #63: Commit c57f47b pushed by BrutalCaeser
19s main
main
19s
experiments: DP results -- both overfit hard, my prediction was wrong ci #62: Commit 03b4573 pushed by BrutalCaeser
24s main
main
24s
experiments: pre-register the DP encoder A/B before any result lands ci #61: Commit 642f7dd pushed by BrutalCaeser
22s main
main
22s
hpc: size the DP run from measurement -- batch 64 + 32 workers, 6.82 h ci #60: Commit 3a68e10 pushed by BrutalCaeser
22s main
main
22s
hpc: install the diffusion extra, and fix a guard that could never pass ci #59: Commit ec1509a pushed by BrutalCaeser
15s main
main
15s
deployment: file the measured control-rate budget; add the DP trainin… ci #58: Commit 38657b8 pushed by BrutalCaeser
26s main
main
26s
hpc: calibration job to size the DP run from a measurement, not my ar… ci #57: Commit 88b7b19 pushed by BrutalCaeser
20s main
main
20s
docs: the base clamp position is part of the frame, and nothing recor… ci #56: Commit bfae145 pushed by BrutalCaeser
16s main
main
16s
utils: replay_compare -- replay an episode in Rerun, and READ THE ARM… ci #55: Commit b03155f pushed by BrutalCaeser
17s main
main
17s
CORRECTION: rollouts are not zero -- 17 are already scored ci #54: Commit 395a042 pushed by BrutalCaeser
20s main
main
20s
watch_rollouts: fail fast on a bad port, and stop two silent hangs ci #53: Commit e9ab821 pushed by BrutalCaeser
24s main
main
24s
utils: infer_offline -- inference on recorded episodes, in degrees ci #52: Commit 988e7ea pushed by BrutalCaeser
16s main
main
16s
utils: watch_rollouts -- step through rollouts one at a time, no scoring ci #51: Commit 9d6a4a8 pushed by BrutalCaeser
18s main
main
18s
models: publish both recovery ACT checkpoints, with the results read … ci #50: Commit b49abac pushed by BrutalCaeser
19s main
main
19s
train: ACT on the recovery dataset, image augmentation A/B ci #49: Commit ce60171 pushed by BrutalCaeser
24s main
main
24s