Views
No views yet
object finetune of
nvidia/GR00T-N1.5-3B for
inference with vla.cpp, a lightweight
C++ inference engine for Vision-Language-Action models built on top of
llama.cpp.lerobot/eagle2hg-processor-groot-n1p5)
is pulled from the Hub by the client, so it is not bundled in this repo.| File | Size | Description |
|---|---|---|
gr00tn1d5-libero-object.gguf | 6.46 GiB | Combined VLA model — Eagle-2 backbone + flow-matching action head + arch config, BF16 |
dataset_statistics.json | — | Action/state normalisation stats (required by the client) |
1# Terminal 1 — serve (use the CUDA build for inference). No mmproj argument.
2VLA_GR00T_BF16_WEIGHTS=1 VLA_GR00T_EMBODIMENT=new_embodiment \
3 ./build-cuda/vla-server --bind tcp://*:5566 \
4 gr00tn1d5-libero-object.gguf
5
6# Terminal 2 — drive a LIBERO episode (inside the LIBERO uv venv)
7python eval/client/run_sim_client_direct.py \
8 --arch gr00t_n1_5 \
9 --task libero_object --task-id 0 --n-episodes 10 \
10 --stats-json dataset_statistics.json \
11 --vla-addr tcp://localhost:5566VLA_GR00T_EMBODIMENT=new_embodiment and VLA_GR00T_BF16_WEIGHTS=1 (the
latter is needed to fit an 8 GB card).lerobot/eagle2hg-processor-groot-n1p5 tokenizer
from the Hub (trust_remote_code); no --tokenizer is needed.--stats-json dataset_statistics.json (action/state un-normalisation).--n-action-steps 16 for this checkpoint.libero_object sweep (10 tasks × 20 episodes = 200 episodes):| Hardware | n_act | Success rate | client/step | client/call | Peak mem |
|---|---|---|---|---|---|
| RTX 3060 (sm_86) | 16 | 96.0% | 14.17 ms | 227 ms | 4866 MiB VRAM |
| Jetson AGX Orin (sm_87) | 16 | 97.5% | 28.78 ms | 461 ms | 1331 MiB RAM |
| Jetson Orin Nano 8 GB (sm_87) | 16 | 96.0% | 84.76 ms | 1356 ms | 4399 MiB sys-Δ |
Unlike GR00T-N1.6 / N1.7, N1.5's footprint fits the Jetson Orin Nano 8 GB. The Nano row was run split (server on the Nano, LIBERO client on a separate host);sys-Δis the system-used-RAM rise, the faithful unified-memory figure on Tegra.
nvidia/GR00T-N1.5-3B
(NVIDIA license — review and accept it before use). The vla.cpp conversion tooling
and inference engine are Apache-2.0-licensed.