# trtruntime
`trtruntime` is a lightweight Python package that provides a TensorRT inference runtime similar in API style to [onnxruntime](https://onnxruntime.ai/). It allows easy loading and running of TensorRT engines with a clean and simple interface.
## Features
- Load serialized TensorRT engine files (`*.engine`) or plan files (`*.plan`)
- Automatically handle CUDA memory bindings and streams
- Simple API modeled after onnxruntime's `InferenceSession`
- Supports querying input and output tensor metadata
- Compatible with TensorRT, PyCUDA, and NumPy
## Installation
```bash
pip install trtruntime
````
> **Note:** You need to have TensorRT and PyCUDA installed on your system.
## Usage
```python
import numpy as np
from trtruntime import InferenceSession
# Create inference session with your TensorRT engine file
sess = InferenceSession("model.engine")
# Prepare input feed as dictionary {input_name: numpy_array}
input_feed = {
"input_1": np.random.rand(1, 3, 224, 224).astype(np.float32),
}
# Run inference
outputs = sess.run(output_names=None, input_feed=input_feed)
# outputs is a list of numpy arrays corresponding to requested outputs
print(outputs)
```
Raw data
{
"_id": null,
"home_page": null,
"name": "trtruntime",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": null,
"keywords": "deepstream, gstreamer, cli, onnx, automation, ai, artificial intelligence, computer vision, edge ai, intelligent edge, craftifai, intelligent edge systems, ies, nvidia, jetson, inference, opencv, triton inference server, tensorrt, onnxruntime",
"author": null,
"author_email": "Mohd Yusuf <yusuf.intelligentedgesystems@gmail.com>",
"download_url": "https://files.pythonhosted.org/packages/c6/e7/c788104090a4ef314dce751dca3f9f2ee3b11cc04ca410cb717c82c7bd91/trtruntime-0.1.0.tar.gz",
"platform": null,
"description": "# trtruntime\n\n`trtruntime` is a lightweight Python package that provides a TensorRT inference runtime similar in API style to [onnxruntime](https://onnxruntime.ai/). It allows easy loading and running of TensorRT engines with a clean and simple interface.\n\n## Features\n\n- Load serialized TensorRT engine files (`*.engine`) or plan files (`*.plan`)\n- Automatically handle CUDA memory bindings and streams\n- Simple API modeled after onnxruntime's `InferenceSession`\n- Supports querying input and output tensor metadata\n- Compatible with TensorRT, PyCUDA, and NumPy\n\n## Installation\n\n```bash\npip install trtruntime\n````\n\n> **Note:** You need to have TensorRT and PyCUDA installed on your system.\n\n## Usage\n\n```python\nimport numpy as np\nfrom trtruntime import InferenceSession\n\n# Create inference session with your TensorRT engine file\nsess = InferenceSession(\"model.engine\")\n\n# Prepare input feed as dictionary {input_name: numpy_array}\ninput_feed = {\n \"input_1\": np.random.rand(1, 3, 224, 224).astype(np.float32),\n}\n\n# Run inference\noutputs = sess.run(output_names=None, input_feed=input_feed)\n\n# outputs is a list of numpy arrays corresponding to requested outputs\nprint(outputs)\n```\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "A lightweight TensorRT inference runtime for Python, inspired by onnxruntime",
"version": "0.1.0",
"project_urls": null,
"split_keywords": [
"deepstream",
" gstreamer",
" cli",
" onnx",
" automation",
" ai",
" artificial intelligence",
" computer vision",
" edge ai",
" intelligent edge",
" craftifai",
" intelligent edge systems",
" ies",
" nvidia",
" jetson",
" inference",
" opencv",
" triton inference server",
" tensorrt",
" onnxruntime"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "05fdcace1809edc4f6a8c063b2e3c18faa1bf4e31b83b017c555e1848cb8e82e",
"md5": "43989a734bd82ac3e57b84b1e0c423c9",
"sha256": "6d76c7a4c3848f5665d8e499163ffad7ef23b4bc3fd17db414dc4379631ebce9"
},
"downloads": -1,
"filename": "trtruntime-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "43989a734bd82ac3e57b84b1e0c423c9",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 3298,
"upload_time": "2025-08-10T20:12:57",
"upload_time_iso_8601": "2025-08-10T20:12:57.483995Z",
"url": "https://files.pythonhosted.org/packages/05/fd/cace1809edc4f6a8c063b2e3c18faa1bf4e31b83b017c555e1848cb8e82e/trtruntime-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "c6e7c788104090a4ef314dce751dca3f9f2ee3b11cc04ca410cb717c82c7bd91",
"md5": "604581c86d9cc0056052b8cdc6444b2e",
"sha256": "976afbc3e5789fca0166ed3ef21f6fc7ecbae30771f6f43a5e23c0d04f83d55d"
},
"downloads": -1,
"filename": "trtruntime-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "604581c86d9cc0056052b8cdc6444b2e",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 3136,
"upload_time": "2025-08-10T20:12:58",
"upload_time_iso_8601": "2025-08-10T20:12:58.813009Z",
"url": "https://files.pythonhosted.org/packages/c6/e7/c788104090a4ef314dce751dca3f9f2ee3b11cc04ca410cb717c82c7bd91/trtruntime-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-08-10 20:12:58",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "trtruntime"
}
JSON