(feat) add XAI ChatCompletion Support

[ishaan-jaff]

Add XAI chatCompletion Support

Checklist before merging:

• Add unit tests for xai/chat/xai_transformation.py
• Add a test for xai streaming

XAI

https://docs.x.ai/docs

:::tip

We support ALL XAI models, just set model=xai/<any-model-on-xai> as a prefix when sending litellm requests

:::

API Key

# env variable
os.environ['XAI_API_KEY']

Sample Usage

from litellm import completion
import os

os.environ['XAI_API_KEY'] = ""
response = completion(
    model="xai/grok-beta",
    messages=[
        {
            "role": "user",
            "content": "What's the weather like in Boston today in Fahrenheit?",
        }
    ],
    max_tokens=10,
    response_format={ "type": "json_object" },
    seed=123,
    stop=["\n\n"],
    temperature=0.2,
    top_p=0.9,
    tool_choice="auto",
    tools=[],
    user="user",
)
print(response)

Sample Usage - Streaming

from litellm import completion
import os

os.environ['XAI_API_KEY'] = ""
response = completion(
    model="xai/grok-beta",
    messages=[
        {
            "role": "user",
            "content": "What's the weather like in Boston today in Fahrenheit?",
        }
    ],
    stream=True,
    max_tokens=10,
    response_format={ "type": "json_object" },
    seed=123,
    stop=["\n\n"],
    temperature=0.2,
    top_p=0.9,
    tool_choice="auto",
    tools=[],
    user="user",
)

for chunk in response:
    print(chunk)

Usage with LiteLLM Proxy Server

Here's how to call a XAI model with the LiteLLM Proxy Server

1. Modify the config.yaml

model_list:
  - model_name: my-model
    litellm_params:
      model: xai/<your-model-name>  # add xai/ prefix to route as XAI provider
      api_key: api-key                 # api key to send your model

2. Start the proxy

$ litellm --config /path/to/config.yaml

3. Send Request to LiteLLM Proxy Server

import openai
client = openai.OpenAI(
    api_key="sk-1234",             # pass litellm proxy key, if you're using virtual keys
    base_url="http://0.0.0.0:4000" # litellm-proxy-base url
)

response = client.chat.completions.create(
    model="my-model",
    messages = [
        {
            "role": "user",
            "content": "what llm are you"
        }
    ],
)

print(response)

curl --location 'http://0.0.0.0:4000/chat/completions' \
    --header 'Authorization: Bearer sk-1234' \
    --header 'Content-Type: application/json' \
    --data '{
    "model": "my-model",
    "messages": [
        {
        "role": "user",
        "content": "what llm are you"
        }
    ],
}'

Relevant issues

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[REQUIRED] Testing - Attach a screenshot of any new tests passing locall

If UI changes, send a screenshot/GIF of working UI fixes

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Nov 1, 2024 1:29pm

[Manouchehri]

Any progress on this? :)

[edmundloo]

@ishaan-jaff: Are there any updates here or anything I can help with? I would love to see this change launched. Many engineers I know, including me, are eagerly waiting for Grok support on litellm.

[ishaan-jaff]

working on this @edmundloo @Manouchehri - hoping to merge today

To fix the problem, we should avoid logging the entire response object directly. Instead, we can log only the non-sensitive parts of the response. This can be achieved by extracting and logging only the necessary information that does not include sensitive data.

• Identify the lines where the response object is being printed.
• Replace the direct logging of the response object with logging of non-sensitive parts of the response.
• Ensure that no sensitive information is included in the logs.

To fix the problem, we should avoid printing sensitive data directly. Instead, we can log a sanitized version of the data or use a logging mechanism that redacts sensitive information. In this case, we will replace the print(chunk) statement with a logging statement that ensures sensitive information is not exposed.

• Replace the print(chunk) statement with a logging statement that redacts sensitive information.
• Ensure that the logging mechanism used is capable of handling sensitive data appropriately.

[codecov]

Codecov Report

Attention: Patch coverage is 88.57143% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/utils.py	60.00%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

[mehmetalpayy]

It's still not available right?