mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-03 02:22:24 +00:00
- measure every file part and every embedded image field the request carries instead of the caller's image arg, so a transform that drops, filters, or adds parts and a JSON body that smuggles extra_body references all bill what the provider actually receives - read seekable streams from offset 0 and restore the cursor, matching how httpx serializes the part - measure input_image* base64 fields on azure image generation the same as edits - merge top-level litellm_params pricing kwargs over nested model_info in _deployment_model_info so custom pricing works on non-router calls - reject bool and non-positive width/height in the size fallback and clamp negative reference_pixels to zero - prices_tokens now checks rates for non-zero values; get_model_info normalizes absent token rates to 0, which made is not None always true and zero-billed per-pixel models that return usage |
||
|---|---|---|
| .. | ||
| image_edit | ||
| image_generation | ||
| passthrough | ||
| realtime | ||
| response | ||
| search | ||
| text_to_speech | ||
| vector_stores | ||
| __init__.py | ||