Builder keeps one host copy of a weight for every layer that reads it transposed
Les mainteneurs répondent en général sous 6 jours
Personne n'a encore pris cette issue.
Évaluation
- Difficulté
- 4/5
- Temps estimé
- 3-5 jours
- Accessibilité débutants
- 25/100
- Type d'issue
- Bug
- Clarté
- Clairement spécifiée
- Activité
- Active
- Domaine
- compilers, performance
Piste de recherche
Start by running python trt_transpose_readers.py one_constant 10 and python trt_transpose_readers.py pretransposed 10 to reproduce the gap in host peak memory. The payload names no source files, so the builder code that handles transposed matrix-multiply weights has to be located. Done means one transposed constant read by N layers keeps host peak close to the pretransposed numbers, about 6.9 GiB for N=10.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Description
What is wrong
When several matrix multiplies read the same weight with MatrixOperation.TRANSPOSE, the builder seems to keep a separate host copy of that weight for each reader. In the script below, one 1 GiB float32 weight read by 10 layers needs 16.1 GiB of host memory to build. Transposed once up front and then read by the same 10 layers, it needs 6.9 GiB. Each extra transposed reader adds about 1 GiB, one more copy of the weight.
Who hits it: any model that runs one block many times in one graph. Examples are a flow-matching or diffusion policy with its denoising loop unrolled, or classifier-free guidance that runs the same layers twice. Linear layers read their weight transposed, so this is the common case. Through Torch-TensorRT, a 3.5 GiB float32 policy that runs its action transformer 10 times needed 20.5 GiB of host memory to compile, and 11.0 GiB once the converter shared one transposed copy itself (pytorch/TensorRT#4965). Related: pytorch/TensorRT#4867.
What this blocks
xvla and eo1 (pytorch/executorch#24353) read the same weights in many layers, so their compile keeps one extra host copy per reader.
Torch-TensorRT 2.15 avoids it only with the opt-in deduplicate_weights (pytorch/TensorRT#4965). Anyone who builds with the TensorRT API, or without that option, still pays it.
A fix in the builder keeps that memory flat however many layers read a weight, which matters most on Jetsons with little memory.
Repro
python trt_transpose_readers.py MODE [N] [D], with MODE one of:
per_reader: N constant layers that point at the same host memory, each read transposed.one_constant: one constant layer, read transposed by N layers.pretransposed: one constant, transposed once with a shuffle layer, read by N layers without a transpose flag.
Default N = 10, D = 16384 (1 GiB float32). TF32 is off.
| mode | N = 1 | N = 5 | N = 10 |
|---|---|---|---|
| per_reader | 16061 MiB (15.7x) | ||
| one_constant | 6918 MiB (6.8x) | 10939 MiB (10.7x) | 16060 MiB (15.7x) |
| pretransposed | 6931 MiB (6.8x) | 6931 MiB (6.8x) | 6930 MiB (6.8x) |
Host peak during the build, weight 1024 MiB, host memory before the build 1603 MiB. The plan is 1024 MiB in every case.
Expected
Host memory that does not grow with the number of readers, as in pretransposed: at most one builder copy of the weight, however many layers read it transposed.
Versions
TensorRT 11.4.0.106, numpy 2.5.4, H100, driver 580.126.09, Linux x86_64.
trt_transpose_readers.py
import sys, threading, time
import numpy as np
import tensorrt as trt
MODE = sys.argv[1]
N = int(sys.argv[2]) if len(sys.argv) > 2 else 10
D = int(sys.argv[3]) if len(sys.argv) > 3 else 16384
peak = [0]
def rss_mib():
for line in open("/proc/self/status"):
if line.startswith("VmRSS"):
return int(line.split()[1]) // 1024
def sampler():
while True:
peak[0] = max(peak[0], rss_mib())
time.sleep(0.05)
w = (np.random.default_rng(0).standard_normal((D, D)) / np.sqrt(D)).astype(np.float32)
w_mib = w.nbytes >> 20
builder = trt.Builder(trt.Logger(trt.Logger.WARNING))
net = builder.create_network(0)
x = net.add_input("x", trt.float32, (8, D))
def constant():
return net.add_constant((D, D), trt.Weights(w)).get_output(0)
shared = constant() if MODE in ("one_constant", "pretransposed") else None
if MODE == "pretransposed":
t = net.add_shuffle(shared)
t.first_transpose = (1, 0)
shared = t.get_output(0)
op = trt.MatrixOperation.NONE if MODE == "pretransposed" else trt.MatrixOperation.TRANSPOSE
for _ in range(N):
weight = shared if shared is not None else constant()
y = net.add_matrix_multiply(x, trt.MatrixOperation.NONE, weight, op).get_output(0)
x = net.add_activation(y, trt.ActivationType.TANH).get_output(0)
net.mark_output(x)
cfg = builder.create_builder_config()
cfg.clear_flag(trt.BuilderFlag.TF32)
start = rss_mib()
threading.Thread(target=sampler, daemon=True).start()
plan = builder.build_serialized_network(net, cfg)
time.sleep(0.2)
print(f"{MODE}: N={N}, weight {w_mib} MiB, host before build {start} MiB, "
f"build peak {peak[0]} MiB ({peak[0] / w_mib:.1f}x the weight), plan {plan.nbytes >> 20} MiB")
Actual
per_reader: N=10, weight 1024 MiB, host before build 1603 MiB, build peak 16061 MiB (15.7x the weight), plan 1024 MiB
one_constant: N=10, weight 1024 MiB, host before build 1603 MiB, build peak 16060 MiB (15.7x the weight), plan 1024 MiB
one_constant: N=5, weight 1024 MiB, host before build 1603 MiB, build peak 10939 MiB (10.7x the weight), plan 1024 MiB
one_constant: N=1, weight 1024 MiB, host before build 1603 MiB, build peak 6918 MiB (6.8x the weight), plan 1024 MiB
pretransposed: N=10, weight 1024 MiB, host before build 1603 MiB, build peak 6930 MiB (6.8x the weight), plan 1024 MiB
pretransposed: N=5, weight 1024 MiB, host before build 1603 MiB, build peak 6931 MiB (6.8x the weight), plan 1024 MiB
pretransposed: N=1, weight 1024 MiB, host before build 1603 MiB, build peak 6931 MiB (6.8x the weight), plan 1024 MiB
- Langage dominant
- C++
- Étoiles
- 13.5k
- Forks
- 2.4k
- Merge moyen
- 6 j 21 h
- PR mergées (30 j)
- 3
Préparer son environnement
- Aucun Dockerfile ni fichier Docker Compose
- Aucun modèle de pull request
- Lire le guide de contribution
Par où commencer
- Lisez l'issue en entier, puis le guide de contribution du projet.
- Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
- Forkez le dépôt et travaillez sur une branche.
- Ouvrez une pull request qui référence le numéro de l'issue.
Autres issues de NVIDIA/TensorRT
-
Module:Polygraphy
Difficulté 2/5 1-3 heures Accessibilité débutants 77/100
Les mainteneurs répondent en général sous 6 jours
-
Polygraphy NumPy Top-K sorts unsigned and minimum signed integers incorrectlyPeut-être pris @0z5a l’a pris il y a 16 jours. OuverteModule:Polygraphy
Difficulté 2/5 1-3 heures Accessibilité débutants 72/100
Les mainteneurs répondent en général sous 6 jours
-
Polygraphy INT32 comparison can overflow and falsely pass at zero tolerancePeut-être pris @0z5a l’a pris il y a 16 jours. OuverteModule:Polygraphy
Difficulté 2/5 1-3 heures Accessibilité débutants 72/100
Les mainteneurs répondent en général sous 6 jours
-
onnx-graphsurgeon: Graph.toposort() hits RecursionError when a graph has a local function and an If/Loop nodePeut-être pris @gabortar l’a pris il y a 39 jours. OuverteModule:GraphSurgeon
Difficulté 2/5 1-3 heures Accessibilité débutants 82/100
Les mainteneurs répondent en général sous 6 jours
-
Polygraphy parse_meta accepts a shape with a missing bracketPeut-être pris @SashaMIT l’a pris il y a 51 jours. OuverteModule:Polygraphy
Difficulté 2/5 1-3 heures Accessibilité débutants 86/100
Les mainteneurs répondent en général sous 6 jours
Toutes les issues de NVIDIA/TensorRT
Issues similaires
-
Difficulté 1/5 Moins d'une heure Accessibilité débutants 91/100
Les mainteneurs répondent en général sous 1 jour
-
Difficulté 2/5 1-3 heures Accessibilité débutants 69/100
maplibre/maplibre-native#5072 ·
Les mainteneurs répondent en général sous 1 jour
-
[UBSan] spawn_simulator passes null child_options.args to uv_spawn (execvp nonnull violation)Ouverte
Difficulté 2/5 1-3 heures Accessibilité débutants 79/100
tenstorrent/tt-umd#4017 ·
Les mainteneurs répondent en général sous 3 jours
-
difficulty:easy func:gui prio:low type:bug
Difficulté 2/5 1-3 heures Accessibilité débutants 72/100
NatronGitHub/Natron#1538 ·
Les mainteneurs répondent en général sous 1 jour
-
Difficulté 2/5 1-3 heures Accessibilité débutants 72/100
hyprwm/hyprsunset#103 · 1 commentaire ·