Pin PyTorch and fix tests

update paths

fix LIBERO config

small fix

Fix conversion script

Fix lint errors

remove ema, remove redundant compute for data loading, memory opt for float32, and set float32 by default for finetuning

add pytorch support for openpi

DROID fix

DROID training

Fix lora ffn

Default weight decay back to 1e-10

fix droid action space for pi05 finetuning

Default weight decay to 0

LIBERO updates

add droid lerobot doc, custom finetuning example

Add pi05 aloha configs

Add Pi05 LIBERO config

Minor nits

Make it harder to forget quantile norm

Add missing use_quantile_norm

Default max_token_len=200 for pi05

Add Pi05-LIBERO config and update README with Pi05 info

add pi05 droid config

Fix normalize/padding ordering

Pi05 support

Fix dataset repeat

make new droid idle filter default

Merge branch 'main' into main

minor edits, requested by karl

subsume filter_last_n_in_ranges into json ranges

Fix normalize tests and add multi-batch dimension test

fix precommit formatting issues

update to follow required merge patterns

remove old filter

update comments, readme

move filter_last_n_in_ranges to json

move option to remove last n in filter ranges to json generator

fix minor bugs

add json filtering

[fix] optimise `_merge_params` to prevent CPU overload (#616)

Fix compute_norm_stats batching

Merge branch 'main' into optimise-merge-params

introduce simple optimisations

fix

move filtering idle actions after flatten

gemma_300m_lora variant is missing in Variant (#555)

Merge branch 'main' into patch-2

Fix array_typing breakage in REPL

Apply suggestions from code review