Search public pages, research tools, and SAVRN solutions.
LoRA fine-tune của π0.5 (pi05base) cho tay máy Yahboom X3Plus (5 khớp + gripper, 20 Hz, 2 camera), nhiệm vụ "pick up the red cube and put it in the bowl". - Chưa kiểm chứng trên robot thật. Mọi số liệu ở trên là loss huấn luyện.
By Hiếu Hoàng, published under apache-2.0, revision 7e9cc4851a88.
LoRA fine-tune của π0.5 (pi05base) cho tay máy Yahboom X3Plus (5 khớp + gripper, 20 Hz, 2 camera), nhiệm vụ "pick up the red cube and put it in the bowl". - Chưa kiểm chứng trên robot thật. Mọi số liệu ở trên là loss huấn luyện. - Loss đi ngang từ bước ~4.000. 6.000 bước sau chỉ giảm thêm 36%, trong biên độ nhiễu. - Một nhiệm vụ, một bối cảnh, một bộ camera. Nhiều khả năng hỏng khi đổi vị trí - Camera cổ tay (USB webcam) cho ảnh mờ, nhiều frame gần như trắng khi áp sát mặt bàn. Cần config pi05x3pluslora và lớp LeRobotX3PlusDataConfig tương ứng trong openpi (ánh xạ astrargb → base0rgb, usbcam → leftwrist0rgb, delta mask (5, -1)). yahboomx3plus · joint1–5 tính bằng radian (URDF x3plusarm)…
LoRA fine-tune của π0.5 (pi05_base)
cho tay máy Yahboom X3Plus (5 khớp + gripper, 20 Hz, 2 camera), nhiệm vụ
"pick up the red cube and put it in the bowl".
| Mô hình nền | gs://openpi-assets/checkpoints/pi05_base |
| Dữ liệu | 115 episode · 29.130 frame · 24,3 phút · một bối cảnh |
| Bước | 10.000 (≈5,5 epoch) · batch 16 |
| Thời gian | 4 giờ 02 phút trên RTX 5090 (32 GB) |
| Loss | 0,1143 → 0,0050 |
action_horizon |
10 |
discrete_state_input |
True |
| Không gian action | 5 khớp delta + gripper tuyệt đối |
| Nhóm | Số lượng | |
|---|---|---|
| LoRA adapter (hạng 16 / 32) | 49.987.584 | train |
| SigLIP vision encoder | 414.803.696 | train đầy đủ |
Projection (action_in/out_proj, time_mlp) |
2.165.792 | train |
| LLM PaliGemma | 2.936.464.384 | đóng băng |
| Tổng train | 466.957.072 (13,72%) |
params/ trọng số (6,0 GB)
train_state/ trạng thái optimizer, để resume training (3,0 GB)
assets/x3plus/pick_red_cube/norm_stats.json thống kê chuẩn hóa — BẮT BUỘC
huggingface-cli download hoanghieu16720/pi05-x3plus-lora --local-dir ./ckpt
cd openpi
export OPENPI_DATA_HOME=/path/to/openpi_data
.venv/bin/python scripts/serve_policy.py policy:checkpoint \
--policy.config=pi05_x3plus_lora \
--policy.dir=./ckpt
Cần config pi05_x3plus_lora và lớp LeRobotX3PlusDataConfig tương ứng trong openpi
(ánh xạ astra_rgb → base_0_rgb, usb_cam → left_wrist_0_rgb, delta mask (5, -1)).
yahboom_x3plus · joint1–5 tính bằng radian (URDF x3plus_arm), gripper 0 = mở, 1 = đóng.
46 files, 9.5 GB in total.
| File | Type | Size | SHA-256 |
|---|---|---|---|
| assets/x3plus/pick_red_cube/norm_stats.json | Configuration | 1.6 KB | — |
| code/apply_patch.py | Configuration | 5.8 KB | — |
| code/serve_x3plus.py | Configuration | 6.9 KB | — |
| code/x3plus_policy.py | Configuration | 2.9 KB | — |
| README.md | Documentation | 2.5 KB | — |
| USAGE.md | Documentation | 3.6 KB | — |
| _CHECKPOINT_METADATA | Other | 426 B | — |
| params/_METADATA | Other | 33.1 KB | — |
| params/_sharding | Other | 17.8 KB | — |
| params/array_metadatas/process_0 | Other | 12.9 KB | — |
| params/d/c3fcae0931e8bb70c30572e9c1c8fc10 | Other | 2.6 KB | — |
| params/manifest.ocdbt | Other | 118 B | — |
| params/ocdbt.process_0/d/06a87e720e210d5f5ffbe32b9c84c342 | Other | 828.9 MB | 717a42d6fd6b |
| params/ocdbt.process_0/d/0fbfea0ac05acf9d2ce4ee9d169bbee5 | Other | 298.1 MB | 306287cf8529 |
| params/ocdbt.process_0/d/1b0142667457b1dd3949a3e5d10d0be7 | Other | 1.2 GB | 0bcfefc5db04 |
| params/ocdbt.process_0/d/23404baffb6fb91dd955bd49895aca02 | Other | 1.9 GB | 61847b9c5c8c |
| params/ocdbt.process_0/d/2dcee75323677ac56ab1910bc823670d | Other | 3.9 MB | 5ba0b2efb879 |
| params/ocdbt.process_0/d/4559703d9940f536b459a8b60554a771 | Other | 1.4 KB | — |
| params/ocdbt.process_0/d/8c9dbc7e294bd6bac9f4d6fa23136470 | Other | 99.0 MB | b441eb10110c |
| params/ocdbt.process_0/d/a209f832283ae9897d9f57c8d1aea180 | Other | 5.7 KB | — |
| params/ocdbt.process_0/d/a8b5f10e202aaf3a945e9d722655939f | Other | 213 B | — |
| params/ocdbt.process_0/d/d893b89e3297da7a24ca3a621bd0f454 | Other | 152.1 MB | 7ecb81b9f8cc |
| params/ocdbt.process_0/d/dfd58405d70cd2e021f6978ae996812e | Other | 886.9 MB | 647d984ed00d |
| params/ocdbt.process_0/d/e65711f7103e04a7b25f83ccc25da683 | Other | 945.9 MB | 28fe59c791cc |
| params/ocdbt.process_0/manifest.ocdbt | Other | 634 B | — |
| train_state/_METADATA | Other | 60.3 KB | — |
| train_state/_sharding | Other | 27.7 KB | — |
| train_state/array_metadatas/process_0 | Other | 19.9 KB | — |
| train_state/d/ea5cd5f6e91b6067fced6be3aa71942f | Other | 2.7 KB | — |
| train_state/manifest.ocdbt | Other | 118 B | — |
| train_state/ocdbt.process_0/d/32cfc852126da94d1095909c3597cbaf | Other | 198 B | — |
| train_state/ocdbt.process_0/d/46b978cebe34877df4dce16070916c9a | Other | 804 B | — |
| train_state/ocdbt.process_0/d/545d155b07db2dfeca339d3ee264448f | Other | 885.9 MB | 40d3eda0f5ba |
| train_state/ocdbt.process_0/d/56539b4a373481691085b32ea1b3f979 | Other | 1.0 GB | 9046652c163a |
| train_state/ocdbt.process_0/d/72b54adb33c92c3343fde2927376fe48 | Other | 171 B | — |
| train_state/ocdbt.process_0/d/9d57507c3e958780ffeef7db01d53f36 | Other | 130.2 KB | f8c34d6fa2c4 |
| train_state/ocdbt.process_0/d/9f4e18cecda1b2c7a65e471e6613f565 | Other | 791 B | — |
| train_state/ocdbt.process_0/d/ad8f8f6b85d99b460c22d8fc2634d7fa | Other | 1.2 KB | — |
| train_state/ocdbt.process_0/d/bd29df87ce6994771992c56c3119dabb | Other | 756 B | — |
| train_state/ocdbt.process_0/d/c1b09583df312a6fa58f21ac1ea92625 | Other | 85.3 MB | 793f5ce80eb7 |
| train_state/ocdbt.process_0/d/d5d3ba237ed39a4d2653eb98a446f8f1 | Other | 256.2 MB | 0550a5b6a4a6 |
| train_state/ocdbt.process_0/d/d6fdfef6eca4a3604d447dc21b60ee54 | Other | 557 B | — |
| train_state/ocdbt.process_0/d/f41d09820fa8e18664d28bf52aef4982 | Other | 918.6 MB | 2e6f5cc07547 |
| train_state/ocdbt.process_0/d/f811686b2cc1405e6c0537327acf86a7 | Other | 369 B | — |
| train_state/ocdbt.process_0/manifest.ocdbt | Other | 642 B | — |
| .gitattributes | Repository | 3.0 KB | — |
Released by Hiếu Hoàng through its official repository on Hugging Face. Read the license.
Yes. pi05-x3plus-lora is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
LIBERO 4in1（liberospatial / liberoobject / liberogoal / libero10）共 53.19 GB 的 Wan2.2-VAE 编码 latent 缓存： 训练 LIBERO policy（action head / VLA）时直接读取 latent 缓存，避免重复 VAE 编码。 - 窗口模式: windowed（--windowed） - 输出:.pt 文件，每 episode 一个 - dataset（推荐）: https://huggingface.co/datasets/MangoGoes/libero4in1wan2.2vaelatentdataset - model（本仓库）: https://huggingface.co/MangoGoes/libero4in1wan2.2vaelatentcosmosstyle
Wan2.2-TI2V-5B video DiT + 48-joint action head, trainingmode=joint. The base is suhyeok's finalized B-method recipe: a teacher-forced (sigma=0.25) self-EMA teacher plus an iBOT prototype loss at L18 L18, gamma=0.01, two-view. On top of it the 3 PAST cond latent frames are pooled into one motion frame before a chosen block. These are NOT the surrogate ctxpool runs. The surrogate line (older base, pd8 x GA1) lives in hmkang/wamctxpoolxattn and hmkang/wamctxpoolavg. Do not compare across the two sets. Geometry: 4-latin (numframesin=25, numframesout=41, fdf 2) = 4 cond + 2 future latent slots, 96 tokens per latent frame, 576 tokens per row. Effective batch 16 clips x GA 2 x 2 views = 64 rows…
weighted/imatrix quants of https://huggingface.co/DeepCybo/PhysBrain1.5-8B For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/PhysBrain1.5-8B-GGUF This is a vision model - mmproj files (if any) will be in the static repository. If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality quant And here are Artefact2's thoughts on the matter…
This repository contains checkpoints and evaluation artifacts for GR00T fine-tuning. Each epoch folder is a separate model checkpoint; the repository root is an index. - Same 50 total LIBERO Spatial trajectories for every version (5,971 frames). - Vision encoder, language model, and the full action head/DiT are trainable. - Eight epochs maximum; 125 optimizer updates per epoch. - Every epoch checkpoint is uploaded and hash-verified before local weight eviction. - Every checkpoint is evaluated on all ten Spatial tasks, with 50 fixed initial states per task: 500 rollouts. - Success-rate plots use completed simulator evaluations, not training losses. Checkpoints and evaluations appear as the…
本仓保存 RoboTwin 实测模型、TDAA codec、原始结果和 4,000 个视频。 在 FastWAM 中接入预训练 TDAA version3bin24 编解码器，将 [32,14] 动作块编码为 [8,16] latent。策略在 latent 空间做 flow matching，再解码为 32 步绝对关节动作。decoder 使用任务向量和由已执行动作历史的 DCT24 特征生成的 phase；每个 episode 重置历史。 frozen 控制整个 TDAA codec，FastWAM 策略仍参与训练。联合训练额外加入动作重构与进度预测损失。动作 token 从 32 个变为 8 个，仅表示动作表示压缩；本次没有端到端加速测量。 - TDAA codec 与配置：codec.pt、config.json、metadata.json、datasetstatistics.json、taskembeddings.json，来自 80,000 步 AE。 代码仓的下载工具按固定版本获取文件并校验 SHA-256。在代码仓安装环境后运行： 权重保存到 checkpoints/released/{official,tdaafrozen}/step002725.pt。Wan VAE/T5/tokenizer 等基础模型、训练数据和 RoboTwin 仿真资产需另外准备，详见 GitHub 复现说明。本次不包含 optimizer/scheduler 完整训练状态。 以下为 FastWAM 基线与 TDAA Frozen=True 两组实测训练的共同设置，已与各自保存的…
LeRobot policy checkpoints uploaded by goalgen/uploadhfcheckpoints.sh. Each subfolder contains the deployment-ready pretrainedmodel/ payload (model.safetensors + config.json + pre/postprocessor + trainconfig.json).