POST
/
models
/
v2
/
{modelId}
curl --request POST \
  --url https://api.bytez.com/models/v2/{modelId} \
  --header 'Authorization: <api-key>' \
  --header 'Content-Type: application/json' \
  --data '{
  "text": "<string>",
  "messages": [
    {
      "role": "system",
      "content": "<string>"
    }
  ],
  "url": "<string>",
  "base64": "<string>",
  "question": "<string>",
  "context": "<string>",
  "candidate_labels": [
    "<string>"
  ],
  "stream": true,
  "json": true,
  "params": {
    "min_length": 123,
    "max_length": 123,
    "temperature": 123
  }
}'
{
  "error": null,
  "output": "<string>"
}

Authorizations

​
Authorization
string
header
required

Set Authorization header to Key BYTEZ_KEY (e.g. 'Authorization': 'Key ABC123')

Path Parameters

​
modelId
string
required

The model you want to run (e.g., openai-community/gpt2).

Body

application/json
The required inputs vary by task. For text-based tasks like text-generation, summarization, and translation, include the "text" field. For chat tasks, use "messages". For image tasks (like image-to-text, image-classification, object-detection, etc.), supply input using either "url" or "base64". For audio tasks such as automatic-speech-recognition, provide audio using "url" or "base64". Question-based tasks like question-answering require both "question" and "context". Zero-shot tasks need "candidate_labels" and may also require "text" or "image". Some tasks, like unconditional-image-generation, may not need any input. Always refer to the specific model's task to determine the required fields.
​
text
string

Input text to be processed by the model (used for most NLP tasks).

​
messages
object[]

An array of chat messages representing the conversation history. Each message must have a role (system, user, or assistant) and content. The content can be:

  • A simple string - An array of content blocks where each block has a type (e.g. text, image, audio, or video) and the corresponding data via text, url, or base64.
​
url
string

URL to an image, audio, or video file (used for image/audio/video tasks).

​
base64
string

Handles Base64-encoded image, audio, or video. Using URLs is recommended to avoid large payloads, which slow your requests down.

​
question
string

Question text (used in question-answering tasks).

​
context
string

Context paragraph (used in question-answering).

​
candidate_labels
string[]

List of candidate labels (used in zero-shot tasks).

​
stream
boolean

Enable/disable text streaming.

​
json
boolean

Similar to stream, but for media. Streams back media instead returning it in JSON format.

​
params
object

Model-specific parameters. See model documentation for details.

Response

200
application/json
Successful response from the model.
​
error
string

Null if everything is fine

Example:

null

​
output

Model output will either be a string, object, or array of items.