Masked bf16 attention rounds its scores to bf16 on Jetson Thor (sm_110), but not on A100
I maintainer di solito rispondono entro 6 giorni
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 5/5
- Tempo stimato
- Più di una settimana
- Idoneità per principianti
- 8/100
- Tipo di issue
- Bug
- Chiarezza
- Specificata chiaramente
- Stato di attività
- Attiva
- Ambito
- machine-learning
Direzione di ricerca
The repro is a single Python script that compiles one scaled_dot_product_attention with Torch-TensorRT and prints channel 0 for no mask, a bool mask and a float mask. Run it on a Jetson AGX Thor and on an A100 to confirm the masked outputs differ. Done means the masked Thor output matches the float32-score result (+0.4219) and the random-input error falls back to about the PyTorch level, which needs a fix inside TensorRT's attention kernels.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
What is wrong
On Jetson AGX Thor, a bf16 attention layer with a mask (bool or additive float) computes its softmax scores in
bf16. Without a mask it keeps them in float32. On A100 (sm_80), with the same
TensorRT, the masked layer keeps float32 scores, like PyTorch.
When 2 scores are close, bf16 rounding makes them equal, so the softmax weights change. On random inputs the error
grows about 13 times.
What this blocks
On Jetson Thor, evo1 misses its accuracy check and fails (pytorch/executorch#24353).
lingbot_vla_v2 passes there only because LeRobot keeps its attention out of TensorRT on Thor. That splits it into hundreds of pieces and makes it slower than eager PyTorch.
A fix lets both use fused attention on Thor, as they do on the other GPUs.
Repro
The script compiles one scaled_dot_product_attention with Torch-TensorRT, which builds one IAttention layer.
A query and 2 keys are chosen so that their scaled scores (200.45 and 199.55) round to the same bf16 value. With
float32 scores, output channel 0 is +0.42. With bf16 scores it is 0.
import torch, torch.nn.functional as F, torch_tensorrt, tensorrt
B, H, L, S, D = 8, 8, 50, 805, 64
dev = "cuda"
class Attention(torch.nn.Module):
def forward(self, q, k, v, mask=None):
return F.scaled_dot_product_attention(q, k, v, attn_mask=mask)
def near_tie():
q = torch.zeros(B, H, L, D, device=dev); q[..., 0], q[..., 1] = 16.0, 2.0
k = torch.zeros(B, H, S, D, device=dev)
k[..., 0, 0], k[..., 0, 1] = 100.0, 1.796875
k[..., 1, 0], k[..., 1, 1] = 100.0, -1.796875
v = torch.zeros(B, H, S, D, device=dev); v[..., 0, 0], v[..., 1, 0] = 1.0, -1.0
return q.bfloat16(), k.bfloat16(), v.bfloat16()
def random_qkv():
g = torch.Generator(device=dev).manual_seed(0)
return tuple((torch.randn(B, H, n, D, device=dev, generator=g) * s).bfloat16()
for n, s in ((L, 2.0), (S, 3.0), (S, 1.0)))
def reference(q, k, v, mask):
s = q.double() @ k.double().transpose(-2, -1) / D**0.5
if mask is not None:
s = s.masked_fill(~mask, float("-inf")) if mask.dtype == torch.bool else s + mask.double()
return s.softmax(-1) @ v.double()
keep = torch.ones(B, 1, 1, S, dtype=torch.bool, device=dev)
keep[1::2, ..., S - S // 8:] = False # key padding; keys 0 and 1 stay visible
additive = torch.zeros(B, 1, 1, S, device=dev).masked_fill(~keep, float("-inf")).bfloat16().expand(B, H, 1, S).contiguous()
print(torch.cuda.get_device_name(), torch.cuda.get_device_capability(), "TensorRT", tensorrt.__version__)
model = Attention().eval()
for name, mask in (("none", None), ("bool", keep), ("float", additive)):
args = near_tie() + (() if mask is None else (mask,))
trt = torch_tensorrt.dynamo.compile(torch.export.export(model, args), arg_inputs=list(args), min_block_size=1)
with torch.no_grad():
tie = trt(*args)[..., 0].float().mean().item()
rargs = random_qkv() + (() if mask is None else (mask,))
ref = reference(*rargs[:3], mask)
ok = torch.isfinite(ref)
err_trt = (trt(*rargs).double() - ref)[ok].abs().max().item()
err_pt = (model(*rargs).double() - ref)[ok].abs().max().item()
print(f"mask={name:5} channel 0: {tie:+.4f} (float32 scores give +0.4219); "
f"random max error vs float64: TensorRT {err_trt:.4f}, PyTorch {err_pt:.4f}")
Results
| no mask | bool mask | float mask | random error, masked | |
|---|---|---|---|---|
| Thor (11,0) | +0.4219 | +0.0000 | +0.0000 | 0.1206 (PyTorch 0.0090) |
| A100 (8,0) | +0.4219 | +0.4219 | +0.4219 | 0.0106 (PyTorch 0.0106) |
Versions: torch 2.15.0 (CUDA 13.2), torch_tensorrt 2.15.0, TensorRT 11.4.0.106 (all from pip install "torch_tensorrt[executorch]"). Thor: JetPack 7.2.1, driver 595.78, CUDA 13.2.
A100: driver 580.126.09.
Actual
Jetson AGX Thor, driver 595.78:
NVIDIA Thor (11, 0) TensorRT 11.4.0.106
mask=none channel 0: +0.4219 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.0090, PyTorch 0.0084
mask=bool channel 0: +0.0000 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.1206, PyTorch 0.0090
mask=float channel 0: +0.0000 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.1206, PyTorch 0.0090
A100 80 GB, driver 580.126.09:
NVIDIA PG509-210 (8, 0) TensorRT 11.4.0.106
mask=none channel 0: +0.4219 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.0106, PyTorch 0.0106
mask=bool channel 0: +0.4219 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.0106, PyTorch 0.0106
mask=float channel 0: +0.4219 (float32 scores give +0.4219); random max error vs float64: TensorRT 0.0106, PyTorch 0.0106
Expected
Float32 scores for masked attention on Thor, as on A100.
- Lingua principale
- C++
- Stelle
- 13.5k
- Fork
- 2.4k
- Merge medio
- 6g 21h
- PR unite (30g)
- 3
Preparare l'ambiente
- Nessun Dockerfile né file Docker Compose
- Nessun modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di NVIDIA/TensorRT
-
Module:Polygraphy
Difficoltà 2/5 1-3 ore Idoneità per principianti 77/100
I maintainer di solito rispondono entro 6 giorni
-
Polygraphy NumPy Top-K sorts unsigned and minimum signed integers incorrectlyForse già presa @0z5a l’ha presa 16 giorni fa. ApertaModule:Polygraphy
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
I maintainer di solito rispondono entro 6 giorni
-
Polygraphy INT32 comparison can overflow and falsely pass at zero toleranceForse già presa @0z5a l’ha presa 16 giorni fa. ApertaModule:Polygraphy
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
I maintainer di solito rispondono entro 6 giorni
-
onnx-graphsurgeon: Graph.toposort() hits RecursionError when a graph has a local function and an If/Loop nodeForse già presa @gabortar l’ha presa 39 giorni fa. ApertaModule:GraphSurgeon
Difficoltà 2/5 1-3 ore Idoneità per principianti 82/100
I maintainer di solito rispondono entro 6 giorni
-
Polygraphy parse_meta accepts a shape with a missing bracketForse già presa @SashaMIT l’ha presa 51 giorni fa. ApertaModule:Polygraphy
Difficoltà 2/5 1-3 ore Idoneità per principianti 86/100
I maintainer di solito rispondono entro 6 giorni
Tutte le issue di NVIDIA/TensorRT
Issue simili
-
Difficoltà 1/5 Meno di un'ora Idoneità per principianti 91/100
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 69/100
maplibre/maplibre-native#5072 ·
I maintainer di solito rispondono entro 1 giorno
-
[UBSan] spawn_simulator passes null child_options.args to uv_spawn (execvp nonnull violation)Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 79/100
tenstorrent/tt-umd#4017 ·
I maintainer di solito rispondono entro 3 giorni
-
difficulty:easy func:gui prio:low type:bug
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
NatronGitHub/Natron#1538 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
hyprwm/hyprsunset#103 · 1 commento ·