Release Notes for NVIDIA NIM for Object Detection#

This documentation contains the release notes for NVIDIA NIM for Object Detection.

Release 1.5.0#

Summary#

Upgraded to use Triton Inference Server 25.08 to address CVEs.
Added TRT optimized engines for CUDA GPU Compute Capability. Support includes 12.0, 10.0, 9.0, 8.9, 8.6, and 8.0.
Added Triton Ensemble Configuration which supports configuring the underlying Triton Ensemble model pipeline.
Added the NIM_TRITON_PINNED_MEMORY_POOL_MB environment variable.
Added the NIM_TRITON_ENABLE_MODEL_CONTROL environment variable.
Added the NIM_TRITON_IDLE_BYTES_LIMIT environment variable.
Added the NIM_TRITON_FLUSH_INTERVAL environment variable.
Added the NIM_TRITON_RATE_LIMIT environment variable.

Known Issues#

The persistence.enabled value and all related dependent configuration flags are currently non-functional in the NIM helm chart.
If you override NIM_TRITON_MAX_BATCH_SIZE a known TensorRT bug can occur in which a CUDA illegal memory access error is returned in the Triton response. If you must reduce VRAM usage, set NIM_TRITON_OPTIMIZATION_MODE=vram_opt to set the batch size to 1. For details on setting this variable, refer to environment variables.

Release 1.4.0#

Summary#

Add performance optimizations across all object detection NIMs.
Added the NIM_TRITON_CUDA_MEMORY_POOL_MB environment variable.
Added the NIM_REPOSITORY_OVERRIDE environment variable.

Known Issues#

The persistence.enabled value and all related dependent configuration flags are currently non-functional in the NIM helm chart.
If you override NIM_TRITON_MAX_BATCH_SIZE a known TensorRT bug can occur in which a CUDA illegal memory access error is returned in the Triton response. If you must reduce VRAM usage, set NIM_TRITON_OPTIMIZATION_MODE=vram_opt to set the batch size to 1. For details on setting this variable, refer to environment variables.

Release 1.3.0#

Summary#

Added support for B200 GPU.

Known Issues#

The list-model-profiles command incorrectly lists compatible model profiles as incompatible. Select the profile that matches your hardware configuration. This bug does not impact automatic profile selection.

Release 1.2.1#

Summary#

Fixed bug where list-model-profiles command fails to run on hosts that don’t have an NVIDIA GPUs, even when NIM_CPU_ONLY is set.

Known Issues#

The list-model-profiles command incorrectly lists compatible model profiles as incompatible. Select the profile that matches your hardware configuration. This bug does not impact automatic profile selection.

Release 1.2.0#

Summary#

This is a General Availability release of the NVIDIA NIM for Object Detection. This release contains the following changes:

Added support for nemoretriever-page-elements-v2, nemoretriever-graphic-elements-v1, and nemoretriever-table-structure-v1 models.
Added the NIM_TRITON_DYNAMIC_BATCHING_MAX_QUEUE_DELAY_MICROSECONDS and NIM_TRITON_MAX_QUEUE_SIZE environment variables.
You can now use the NIM_TRITON_OPTIMIZATION_MODE environment variable to optimize for performance or VRAM.
Renamed the NIM_TRITON_MODEL_BATCH_SIZE environment variable to NIM_TRITON_MAX_BATCH_SIZE.
Added processed images size (in MB) as usage metric in responses.
Reduced container image sizes.
Removed model profiles for A100 PCIe 40GB & H100 PCIe 80GB configurations.

Object Detection Model Supported#

nemoretriever-page-elements-v2
nemoretriever-graphic-elements-v1
nemoretriever-table-structure-v1

Known Issues#

The list-model-profiles command incorrectly lists compatible model profiles as incompatible. Select the profile that matches your hardware configuration. This bug does not impact automatic profile selection.
The list-model-profiles command fails to run on hosts that don’t have an NVIDIA GPUs, even when NIM_CPU_ONLY is set.

Release 1.1.0-rtx (Beta)#

Summary#

This is a public beta release of the NVIDIA NIM for Object Detection. This release contains the following changes:

Added support for GeForce RTX 4090, NVIDIA RTX 6000 Ada Generation, GeForce RTX 5080, and GeForce RTX 5090 for the nv-yolox-page-elements-v1 NIM.
Added the [NIM_TRITON_MODEL_BATCH_SIZE] environment variable.

Known Issues#

The list-model-profiles command incorrectly lists compatible model profiles as incompatible. Select the profile that matches your hardware configuration. This bug does not impact automatic profile selection.

Release 1.0.0#

Summary#

This is a Early Access release of the NVIDIA NIM for Object Detection. This release contains the following changes:

Updates the /v1/infer endpoint request and response JSON schemas. See the API Reference for more details.

Object Detection Model Supported#

nv-yolox-page-elements-v1

Release 0.2.1#

Summary#

This is an Early Access release of the NVIDIA NIM for Object Detection. This release contains the following fixes:

Fixes bug where NIM_HTTP_TRITON_PORT was not properly setting the port number.
Fixes the /v1/manifest API described in the API Reference returning an empty result instead of the manifest file.
Renames the previously supported model nv-yolox-structured-images-v1 to nv-yolox-page-elements-v1 to better describe the model.

Object Detection Model Supported#

nv-yolox-page-elements-v1

Release 0.2.0#

Summary#

This is the second Early Access release of the NVIDIA NIM for Object Detection. This release contains the following changes:

Added validation for the base64 decoding format and improved the error message for when the format is incorrect.
Improved startup performance of the NIM.
Added FP16 optimized TRT engines for A100, H100, A10G, L40S.

Object Detection Model Supported#

nv-yolox-structured-images-v1

Known Issues#

The /v1/manifest API described in the API Reference returns an empty result instead of the manifest file.

Release 0.1.0#

Summary#

This is the first Early Access release of the NVIDIA NIM for Object Detection.

Object Detection Model Supported#

nv-yolox-structured-images-v1

Known Issues#

If the input image_url does not match the expected format described in the API Reference, the runtime returns an error message, such as {"error": " not enough values to unpack (expected 2, got 1)"}, indicating what formatting error occurred.