DeepInfra embeddings batch with an array, encoding format is float only

TL;DR: For batch embeddings on DeepInfra, pass an array of strings as input in one request instead of looping one request per text. Set encoding_format to float; it is the only supported value, so requests for base64 encodings fail. Models like Qwen/Qwen3-Embedding-8B or BAAI/bge-base-en-v1.5 work through the standard OpenAI embeddings client pointed at the /v1/openai base URL.

Fix

1. For batch embeddings on DeepInfra, pass an array of strings as input in one request instead of looping one request per text. Expected: this specific failure stops.

2. Set encoding_format to float; it is the only supported value, so requests for base64 encodings fail. Expected: this specific failure stops.

Details

Models like Qwen/Qwen3-Embedding-8B or BAAI/bge-base-en-v1.5 work through the standard OpenAI embeddings client pointed at the /v1/openai base URL. Context: DeepInfra docs Embeddings (deepinfra/docs apis/embeddings.mdx): POST YOURHOST.deepinfra.com/v1/openai/embeddings takes model, input (a string or an array of strings), and encodingformat, which is float only. Batch embeddings are a plain array as input. The example uses model Qwen/Qwen3-Embedding-8B via the OpenAI SDK with baseurl YOURHOST.deepinfra.com/v1/openai.

When to use

You hit exactly this: DeepInfra embeddings batch with an array, encoding format is float only.

When not to use

A different error, or the same symptom in a different tool. This page only covers the failure above.

Compatibility

the tool named in the title (versions mentioned here: v1.5, v1).

DeepInfra docs Embeddings (deepinfra/docs apis/embeddings.mdx): POST https://api.deepinfra.com/v1/openai/embeddings takes model, input (a string or an array of strings), and encodingformat, which is float only. Batch embeddings are a plain array as input. The example uses model Qwen/Qwen3-Embedding-8B via the OpenAI SDK with baseurl https://api.deepinfra.com/v1/openai. Resolved: For batch embeddings on DeepInfra, pass an array of strings as input in one request instead of looping one request per text. Set encoding_format to float; it is the only supported value, so requests for base64 encodings fail. Models like Qwen/Qwen3-Embedding-8B or BAAI/bge-base-en-v1.5 work through the standard OpenAI embeddings client pointed at the /v1/openai base URL. Source: https://github.com/deepinfra/docs/blob/HEAD/apis/embeddings.mdx