From f447e9279bc4e77a0e8ecaf3fff5d8801fc224b6 Mon Sep 17 00:00:00 2001 From: Vertax42 <1007644658@qq.com> Date: Wed, 19 Aug 2026 21:04:32 +0800 Subject: [PATCH] docs: drop the README commands whose configs no longer exist MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Checked every `--config-name` / `--policy.config=` / `train.py ` in the README against `all_configs()` - the real resolver, so the five Python-generated RoboArena baselines count as valid, not just the YAML files. Seven names did not resolve, meaning those commands fail at startup for anyone who copies them. Deleted, because the config behind them is gone: - the `..._assemble_box_with_phone_stand_lora_0403` training run (the block keeps its 0422 and 0430 runs, both of which ship) - the `..._shoe_insole_retrieval_and_packing_0607_h100` training run (the block keeps its two 0515 runs) - "BiFlexiv - earbuds case sequential insertion task" (training) and "BiFlexiv - earbuds case assembly with lid operation inference" (deployment) - the whole subsections, since the one config both used is gone - "BiARX5 - training-time RTC inference" - same, `pi05_base_arx5_lora_training_time_rtc` - the `..._assemble_box_with_phone_stand_lora_0410_merged_fixed` serve command (the block keeps three that resolve) Two names were not dead and are corrected rather than dropped: - `..._newbalance_shoe_insole_...` is a *misspelling*. The config ships as `..._newbalacne_...` - the typo is in the filename, and the typo is the one that works. Six call sites, including step ① of the three-machine dewu_video_switch runbook, which is otherwise live. - The getting-started walkthrough used `pi05_base_arx5_lora` as its stand-in name. Deleting a tutorial's commands would leave the tutorial empty, so it now names `pi05_base_xtac_umi_pick_up_cube_0807_h200`, which ships and is the config under active work. After this, every config name in README.md resolves. Left alone: `examples/droid/README_train.md` still names `pi05_droid_finetune` and `pi05_full_droid_finetune`, which this fork does not ship. Deleting those would empty that document's entire Run section, and whether the fork keeps DROID fine-tuning docs at all is a separate call - `pi0_droid`, `pi05_droid` and `pi0_fast_droid` do still ship, so DROID is not simply gone. Co-Authored-By: Claude Opus 5 (1M context) --- README.md | 59 +++++++------------------------------------------------ 1 file changed, 7 insertions(+), 52 deletions(-) diff --git a/README.md b/README.md index 1aba0b8..3723500 100644 --- a/README.md +++ b/README.md @@ -240,13 +240,13 @@ python scripts/dump_config_to_yaml.py --output-dir configs Before we can run training, we need to compute the normalization statistics for the training data. Run the script below with the name of your training config (e.g., for Xense): ```bash -python scripts/compute_norm_stats.py --config-name pi05_base_arx5_lora +python scripts/compute_norm_stats.py --config-name pi05_base_xtac_umi_pick_up_cube_0807_h200 ``` Now we can kick off training with the following command (the `--overwrite` flag is used to overwrite existing checkpoints if you rerun fine-tuning with the same config): ```bash -XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py pi05_base_arx5_lora --exp-name=my_experiment --overwrite +XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py pi05_base_xtac_umi_pick_up_cube_0807_h200 --exp-name=my_experiment --overwrite ``` The command will log training progress to the console and save checkpoints to the `checkpoints` directory. You can also monitor training progress on the Weights & Biases dashboard. For maximally using the GPU memory, set `XLA_PYTHON_CLIENT_MEM_FRACTION=0.9` before running training -- this enables JAX to use up to 90% of the GPU memory (vs. the default of 75%). @@ -258,7 +258,7 @@ The command will log training progress to the console and save checkpoints to th Once training is complete, we can run inference by spinning up a policy server and then querying it from your robot runtime. Launching a model server is easy (we use the checkpoint for iteration 20,000 for this example, modify as needed): ```bash -python scripts/serve_policy.py policy:checkpoint --policy.config=pi05_base_arx5_lora --policy.dir=checkpoints/pi05_base_arx5_lora/my_experiment/19999 +python scripts/serve_policy.py policy:checkpoint --policy.config=pi05_base_xtac_umi_pick_up_cube_0807_h200 --policy.dir=checkpoints/pi05_base_xtac_umi_pick_up_cube_0807_h200/my_experiment/19999 ``` This will spin up a server that listens on port 8000 and waits for observations to be sent to it. We can then run an evaluation script (or robot runtime) that queries the server. @@ -440,11 +440,6 @@ XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ #### BiFlexiv — assemble box with phone stand ```bash -python scripts/compute_norm_stats.py --config-name pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0403 -XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ - pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0403 \ - --exp-name=bi_flexiv_assemble_box_with_phone_stand_lora_20260403 --overwrite - python scripts/compute_norm_stats.py --config-name pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0422_merged_fixed_h100 XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0422_merged_fixed_h100 \ @@ -460,15 +455,6 @@ XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ --exp-name=pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0430_merged_fixed_h100_0510 --overwrite ``` -#### BiFlexiv — earbuds case sequential insertion task - -```bash -python scripts/compute_norm_stats.py --config-name pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100 -XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ - pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100 \ - --exp-name=pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100_0513 --overwrite -``` - #### BiFlexiv - shoe_insole_retrieval_and_packing_0 ```bash @@ -481,11 +467,6 @@ python scripts/compute_norm_stats.py --config-name pi05_base_bi_flexiv_shoe_inso XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ pi05_base_bi_flexiv_shoe_insole_retrieval_and_packing_0515_h100 \ --exp-name=pi05_base_bi_flexiv_shoe_insole_retrieval_and_packing_0515_h100_0519 --overwrite - -python scripts/compute_norm_stats.py --config-name pi05_base_bi_flexiv_shoe_insole_retrieval_and_packing_0607_h100 -XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ - pi05_base_bi_flexiv_shoe_insole_retrieval_and_packing_0607_h100 \ - --exp-name=pi05_base_bi_flexiv_shoe_insole_retrieval_and_packing_0607_h100_0607 --overwrite ``` #### BiFlexiv - shoe_insole_retrieval_and_packing_1 @@ -508,25 +489,9 @@ XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/train.py \ ### Deployment Commands (latest per platform) -#### BiARX5 — training-time RTC inference - -```bash -python scripts/serve_policy.py \ - --default-prompt="pick rgb cubes and place them into the blue box" \ - policy:checkpoint \ - --policy.config=pi05_base_arx5_lora_training_time_rtc \ - --policy.dir=checkpoints/pi05_base_arx5_lora_training_time_rtc/training_time_rtc_20251209/39999 -``` - #### BiFlexiv — assemble box inference ```bash -python scripts/serve_policy.py \ - --default-prompt="assemble the box with the phone stand" \ - policy:checkpoint \ - --policy.config=pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0410_merged_fixed \ - --policy.dir=checkpoints/pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0410_merged_fixed/bi_flexiv_assemble_box_with_phone_stand_lora_0410_merged_fixed_20260413/79999 - python scripts/serve_policy.py \ --default-prompt="assemble the box with the phone stand" \ policy:checkpoint \ @@ -546,16 +511,6 @@ python scripts/serve_policy.py \ --policy.dir=checkpoints/pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0430_merged_fixed_h100/pi05_base_bi_flexiv_assemble_box_with_phone_stand_lora_0430_merged_fixed_h100_0510/66000 ``` -#### BiFlexiv — earbuds case assembly with lid operation inference - -```bash -python scripts/serve_policy.py \ - --default-prompt="Pick up the earbuds from the acrylic plate, open the charging case, precisely align and gently insert the earbuds using contact feedback, then close the lid securely" \ - policy:checkpoint \ - --policy.config=pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100 \ - --policy.dir=checkpoints/pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100/pi05_base_bi_flexiv_earbuds_case_sequential_insertion_teleop_rtc_0513_h100_0513/19999 -``` - #### BiFlexiv - shoe_insole_retrieval_and_packing_0 inference ```bash @@ -580,8 +535,8 @@ python scripts/serve_policy.py \ python scripts/serve_policy.py \ --default-prompt="Take the shoe out of the shoebox, open the shoe tongue, remove and reinsert the insole, then place the shoe into the shoebox." \ policy:checkpoint \ - --policy.config=pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100 \ - --policy.dir=checkpoints/pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100/pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100_0701/77000 + --policy.config=pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100 \ + --policy.dir=checkpoints/pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100/pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100_0701/77000 ``` #### BiFlexiv - bag_inspection_0611 inference @@ -653,8 +608,8 @@ The shoe-insole packing demo runs across three machines: the **inference server* python scripts/serve_policy.py \ --default-prompt="Take the shoe out of the shoebox, open the shoe tongue, remove and reinsert the insole, then place the shoe into the shoebox." \ policy:checkpoint \ - --policy.config=pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100 \ - --policy.dir=checkpoints/pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100/pi05_base_bi_flexiv_newbalance_shoe_insole_retrieval_and_packing_0616_h100_0701/77000 + --policy.config=pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100 \ + --policy.dir=checkpoints/pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100/pi05_base_bi_flexiv_newbalacne_shoe_insole_retrieval_and_packing_0616_h100_0701/77000 ``` **② Screen PC (video-playback laptop) — run on the screen pc side**