API Reference (gRPC) for NVIDIA NIM for Nemotron Parse#
This documentation contains the gRPC reference for NVIDIA NIM for Nemotron Parse.
gRPC Models#
The gRPC model names differ from the NIM model IDs shown in the Support Matrix.
To derive the gRPC model name, replace each slash (/), hyphen (-), and period (.) in the model ID with an underscore (_). The following table shows the mapping for the supported model.
Model ID |
gRPC Model Name |
|---|---|
nvidia/nemotron-3.5-parse |
nvidia_nemotron_3_5_parse |
Request Inputs#
Input |
Shape |
Data Type |
Description |
Required |
|---|---|---|---|---|
|
[1] or [1, 1] |
BYTES |
The raw bytes of one PNG or JPEG page image. |
Yes |
|
[1] or [1, 1] |
BYTES |
UTF-8 |
Yes |
|
[1] or [1, 1] |
BYTES |
UTF-8 JSON containing the same |
No |
Each request contains one page. The image requirements and output-token limit are described in API Usage Examples (OpenAPI) for NVIDIA NIM for Nemotron Parse. Inputs can use typed bytes_contents or KServe raw-input encoding. Raw BYTES values have a four-byte unsigned little-endian length followed by the value. Do not provide both raw and typed contents for the same input.
Request Parameters#
Use the options input for request settings. Request, input, and output parameters and explicit model versions are not supported.
Response#
Output |
Shape |
Data Type |
Description |
|---|---|---|---|
|
[1] |
BYTES |
UTF-8 extracted text with Markdown formatting and coordinate/class control tokens. |
|
[1] |
BYTES |
UTF-8 |
|
[1] |
UINT32 |
Generated output-token count. |
Omit output selection to return all three tensors, or request a subset by name. Outputs use raw_output_contents: BYTES values are length-prefixed, and output_tokens contains one four-byte unsigned little-endian integer.