[Feature]: Support `include_usage` for bedrock

[Manouchehri]

The Feature

It'd be really nice if include_usage worked on providers other than just OpenAI. I think LiteLLM should be able to do this, since we already calculate the cost elsewhere?

Motivation, pitch

It's really useful for users to know how much they've spend in tokens for each streaming request.

Twitter / LinkedIn details

https://www.linkedin.com/in/davidmanouchehri/

[krrishdholakia]

this already works @Manouchehri

e.g. response from predibase.

can you share a case where it didn't work? and we can file that as an issue

[Manouchehri]

It's missing from Bedrock.

curl -v "${OPENAI_API_BASE}/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "claude-3-5-sonnet-20240620",
    "max_tokens": 10,
    "seed": 4242,
    "stream": true,
    "temperature": 0.0,
    "messages": [
      {
        "role": "user",
        "content": "Hello"
      }
    ],
    "stream_options": {
      "include_usage": true
    }
  }'

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":"Hello","role":"assistant"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":"!"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" How"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" can"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" I"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" assist"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" you"}}],"created":1719529524,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" today"}}],"created":1719529525,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":"?"}}],"created":1719529525,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"index":0,"delta":{"content":" Feel"}}],"created":1719529525,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: {"id":"chatcmpl-9569e7af-aaf6-4744-9d8c-0be2bd77f528","choices":[{"finish_reason":"length","index":0,"delta":{}}],"created":1719529525,"model":"anthropic.claude-3-5-sonnet-20240620-v1:0","object":"chat.completion.chunk"}

data: [DONE]

I think it's missing from Azure OpenAI as well, haven't confirmed yet though.

[Manouchehri]

Confirmed, it is also missing/not working for Azure OpenAI requests.

[Manouchehri]

It's missing from Anthropic (directly) too.