Including version in versions.txt

Author: apbose

dev_dep_versions.yml

@@ -1,2 +1,3 @@
 __cuda_version__: "12.8"
 __tensorrt_version__: "10.9.0"
+__tensorrt_llm_version__: "0.17.0.post1"

py/torch_tensorrt/dynamo/utils.py

@@ -20,6 +20,7 @@
 from torch_tensorrt._enums import Platform, dtype
 from torch_tensorrt._features import ENABLED_FEATURES
 from torch_tensorrt._Input import Input
+from torch_tensorrt._version import __tensorrt_llm_version__
 from torch_tensorrt.dynamo import _defaults
 from torch_tensorrt.dynamo._defaults import default_device
 from torch_tensorrt.dynamo._engine_cache import BaseEngineCache
@@ -821,9 +822,8 @@ def download_plugin_lib_path(py_version: str, platform: str) -> str:
     plugin_lib_path = None
 
     # Downloading TRT-LLM lib
-    # TODO: check how to fix the 0.18.0 hardcode below
     base_url = "https://pypi.nvidia.com/tensorrt-llm/"
-    file_name = f"tensorrt_llm-0.18.0-{py_version}-{py_version}-{platform}.whl"
+    file_name = f"tensorrt_llm-{__tensorrt_llm_version__}-{py_version}-{py_version}-{platform}.whl"
     download_url = base_url + file_name
     if not (os.path.exists(file_name)):
         try:
@@ -887,7 +887,7 @@ def load_tensorrt_llm() -> bool:
             return False
         else:
             # this is used as the default py version
-            py_version = "cp312"
+            py_version = "cp310"
             platform = Platform.current_platform()
 
             platform = str(platform).lower()

setup.py

@@ -28,6 +28,7 @@
 __version__: str = "0.0.0"
 __cuda_version__: str = "0.0"
 __tensorrt_version__: str = "0.0"
+__tensorrt_llm_version__: str = "0.0"
 
 LEGACY_BASE_VERSION_SUFFIX_PATTERN = re.compile("a0$")
 
@@ -63,6 +64,7 @@ def get_base_version() -> str:
 def load_dep_info():
     global __cuda_version__
     global __tensorrt_version__
+    global __tensorrt_llm_version__
     with open("dev_dep_versions.yml", "r") as stream:
         versions = yaml.safe_load(stream)
         if (gpu_arch_version := os.environ.get("CU_VERSION")) is not None:
@@ -72,6 +74,7 @@ def load_dep_info():
         else:
             __cuda_version__ = versions["__cuda_version__"]
         __tensorrt_version__ = versions["__tensorrt_version__"]
+        __tensorrt_llm_version__ = versions["__tensorrt_llm_version__"]
 
 
 load_dep_info()
@@ -249,6 +252,7 @@ def gen_version_file():
         f.write('__version__ = "' + __version__ + '"\n')
         f.write('__cuda_version__ = "' + __cuda_version__ + '"\n')
         f.write('__tensorrt_version__ = "' + __tensorrt_version__ + '"\n')
+        f.write('__tensorrt_llm_version__ = "' + __tensorrt_llm_version__ + '"\n')
 
 
 def copy_libtorchtrt(multilinux=False, rt_only=False):