model card: no concrete GPU model names

README.md

@@ -46,7 +46,7 @@ written by Claude (Anthropic) operating autonomously with humans setting goals a
 | Temporal memory | Present in the checkpoint, **baked out** of the export: the deployed model is single-frame |
 | Sparsity | 2:4 structured sparsity on the convolutional trunk (planner branches dense) |
 | Runtime | TensorRT (INT8 on Orin, fp16 elsewhere) via `deploy/runtime.py`, `deploy/orin_realtime.py` or the C++ runtime in `deploy/cpp` |
-| Latency | AGX Orin, INT8, CUDA Graph, zero-copy input: ~70 ms median per 8-camera frame; L40S fp16 (TensorRT 8.6, plugin-free): ~30 ms |
+| Latency | AGX Orin, INT8, CUDA Graph, zero-copy input: ~70 ms median per 8-camera frame; workstation fp16 (TensorRT 8.6, plugin-free, data-center GPU): ~30 ms |
 | Format | PyTorch checkpoint + ONNX (opset 17, no custom ops). TensorRT engines are built locally |
 | License | Apache-2.0 |
 
@@ -102,7 +102,7 @@ hf download AutowareFoundation/meteor-demo-scenes --repo-type dataset --include
 git clone https://github.com/tier4/METEOR && cd METEOR
 python3 hf/onnx_smoke_test.py --onnx ../meteor/meteor_v157c3Z.onnx --root ../demo/valday --frame 40   # -> SMOKE PASS
 
-# TensorRT on a workstation (verified: TensorRT 8.6, L40S, ~10 min build, 194 MB plan)
+# TensorRT on a workstation (verified: TensorRT 8.6, data-center GPU, ~10 min build, 194 MB plan)
 python3 deploy/build_engine_fp16.py ../meteor/meteor_v157c3Z.onnx out/meteor_fp16.engine 8
 METEOR_TH2D=0.30 PYTHONPATH=. python3 deploy/orin_realtime.py --engine out/meteor_fp16.engine \
     --root ../demo/valday --out out/demo.mp4