kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-13 22:18:36 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[None][chore] Lock onnx version <1.20.0 and remove WAR for TRT 10.13 (#9006)

Browse Source 
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
Co-authored-by: Yanchao Lu <yanchaol@nvidia.com>
This commit is contained in:
Yiqing Yan 2025-11-10 10:34:06 +08:00 committed by GitHub
parent 67af7c15a5
commit 78fac1f665
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 60 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
docs/source/installation/linux.md
View File

@ -59,3 +59,15 @@ There are some known limitations when you pip install pre-built TensorRT LLM whe
when OMPI was not configured --with-slurm and we weren't able
to discover a SLURM installation in the usual places.
```
2. Prevent `pip` from replacing existing PyTorch installation
On certain systems, particularly Ubuntu 22.04, users installing TensorRT LLM would find that their existing, CUDA 13.0 compatible PyTorch installation (e.g., `torch==2.9.0+cu130`) was being uninstalled by `pip`. It was then replaced by a CUDA 12.8 version (`torch==2.9.0`), causing the TensorRT LLM installation to be unusable and leading to runtime errors.
The solution is to create a `pip` constraints file, locking `torch` to the currently installed version. Here is an example of how this can be done manually:
```bash
CURRENT_TORCH_VERSION=$(python3 -c "import torch; print(torch.__version__)")
echo "torch==$CURRENT_TORCH_VERSION" > /tmp/torch-constraint.txt
pip3 install --upgrade pip setuptools && pip3 install tensorrt_llm -c /tmp/torch-constraint.txt
```

4
requirements.txt
View File

@ -8,7 +8,7 @@ diffusers>=0.27.0
lark
mpi4py
numpy<2
onnx>=1.18.0
onnx>=1.18.0,<1.20.0
onnx_graphsurgeon>=0.5.2
openai
polygraphy
@ -19,8 +19,6 @@ pandas
h5py==3.12.1
StrEnum
sentencepiece>=0.1.99
# WAR for tensorrt depending on the archived nvidia-cuda-runtime-cu13 package
nvidia-cuda-runtime-cu13==0.0.0a0
tensorrt~=10.13.0
# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-25-10.html#rel-25-10 uses 2.9.0a0.
torch>=2.9.0a0,<=2.9.0

49
tests/unittest/test_pip_install.py
View File

@ -105,6 +105,51 @@ def download_wheel(args):
shell=True)
def install_tensorrt_llm():
"""
Installs the tensorrt_llm wheel, dynamically creating a torch constraint
if torch is already installed to prevent it from being replaced.
"""
print("########## Install tensorrt_llm package ##########")
install_command = "pip3 install tensorrt_llm-*.whl"
# Always check for an existing torch installation, regardless of OS.
try:
print("Checking for existing torch installation...")
torch_version_result = subprocess.run(
['python3', '-c', 'import torch; print(torch.__version__)'],
capture_output=True,
text=True,
check=True)
torch_version = torch_version_result.stdout.strip()
if torch_version:
print(f"Found installed torch version: {torch_version}")
constraint_filename = "torch-constraint.txt"
with open(constraint_filename, "w") as f:
f.write(f"torch=={torch_version}\n")
print(
f"Created {constraint_filename} to constrain torch to version {torch_version}."
)
# Modify install command to use the constraint
install_command += f" -c {constraint_filename}"
else:
# This case is unlikely if the subprocess call succeeds
print(
"Could not determine installed torch version. Installing without constraint."
)
except (subprocess.CalledProcessError, FileNotFoundError):
# This handles cases where python3 fails or 'import torch' raises an error.
print("Torch is not installed. Proceeding without constraint.")
# Execute the final installation command
print(f"Executing command: {install_command}")
subprocess.check_call(install_command, shell=True)
def test_pip_install():
parser = argparse.ArgumentParser(description="Check Pip Install")
parser.add_argument("--wheel_path",
@ -125,8 +170,8 @@ def test_pip_install():
shell=True)
download_wheel(args)
print("########## Install tensorrt_llm package ##########")
subprocess.check_call("pip3 install tensorrt_llm-*.whl", shell=True)
install_tensorrt_llm()
print("########## Test import tensorrt_llm ##########")
subprocess.check_call(
'python3 -c "import tensorrt_llm; print(tensorrt_llm.__version__)"',