More APIs

DEVUP Native API

Direct raw payload forwarding for specialized model architectures.

The DEVUP Native API is a direct pass-through endpoint that forwards raw request payloads to models. It supports three model types: text generation, reranking, and zero-shot image classification.

For standard chat completions and embeddings, the OpenAI-compatible API is recommended. Use the native inference endpoint when you need direct model invocation or model architectures not covered by the OpenAI endpoints.

POSThttps://api.devupai.com/v1/inference/{model_path}

JavaScript client

bash

npm install devupai

Text Generation

const response = await fetch("https://api.devupai.com/v1/inference/deepseek-ai/DeepSeek-V4-Flash-0731", {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${DEVUP_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    inputs: "Hello!",
  }),
});
const data = await response.json();
console.log(data);

Reranking

Pass your query and documents to compute raw relevance scores. Note that queries must be an array even when querying a single string.

curl "https://api.devupai.com/v1/inference/nvidia/llama-nemotron-rerank-vl-1b-v2" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEVUP_API_KEY" \
  -d '{
    "queries": ["what is machine learning"],
    "documents": ["ML is a field of AI", "Paris is in France"]
  }'

The response returns a positional scores array matching the order of the input documents (index 0 scores the first document), where higher values indicate greater relevance:

json

{
  "id": "inf-...",
  "model": "nvidia/llama-nemotron-rerank-vl-1b-v2",
  "scores": [0.3393383026123047, 0.004123545251786709],
  "_devup": { "cost_dzd": 0.0001, "balance_dzd": 4523.64 }
}

For pre-sorted and formatted results, use the dedicated /v1/rerank endpoint instead.

Zero-Shot Image Classification

Zero-shot image classification models are currently unavailable upstream and requests will fail.

bash

curl -X POST \
  -H "Authorization: Bearer $DEVUP_API_KEY" \
  -F "model=openai/clip-vit-large-patch14-336" \
  -F "image=@image.jpg" \
  -F 'candidate_labels=["dog", "cat", "car", "horse", "person"]' \
  'https://api.devupai.com/v1/inference/openai/clip-vit-large-patch14-336'

Other Model Types

Sending a request to /v1/inference/{model} for other model types (such as embeddings, image generation, or speech) returns an HTTP 400 error identifying the dedicated endpoint to use.

HTTP / other languages

The native API is plain HTTP — you can use it from any language (Go, C#, Java, PHP, Ruby, C++, etc.) without any SDK dependency.

Text to Speech