[InferenceClient] Add support for adapter_id (text-generation) and response_format (chat-completion)

[Wauplin]

This PR updates the inference client and types following the latest TGI updates:

• adds adapter_id to text-generation to chose whih LoRA should be loaded (cc @datavistics)
• adds response_format to chat_completion to constraint the response format with either a regex or a jsonschema (cc @aymeric-roucher)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[LysandreJik]

Thanks for the PR @Wauplin, these seem like sensible changes to me!

[Wauplin]

Finally managed to fix the merge conflict + tests 😄 I'll merge now since last test is unrelated.

[aymeric-roucher]

@Wauplin I'm trying this with code:

from huggingface_hub import InferenceClient

client = InferenceClient("meta-llama/Meta-Llama-3.1-70B-Instruct")

client.chat_completion([{"role": "user", "content": "ok"}], response_format={"type": "regex", "value": "*"})

And I get error:

HfHubHTTPError: 424 Client Error: Failed Dependency for url: https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3.1-70B-Instruct/v1/chat/completions (Request ID: _ELWpQiWQT37RR-b07aAr)

Request failed during generation: Server error:

This seems to be due to the regex being invalid. Is it possible to give a more explicit error from the server in case of incorrect regex?

[aymeric-roucher]

Also, constrained generation does not seem to work for me on chat_completion.

Example:

from huggingface_hub import InferenceClient
client = InferenceClient("meta-llama/Meta-Llama-3.1-70B-Instruct")
client.chat_completion([{"role": "user", "content": "ok"}], response_format={"type": "regex", "value": ".+?\nCode"}).choices[0].message.content

Gives output:
"It seems like you're ready to start a conversation. What's on your mind? Want to talk about something specific or ask a question? I'm all ears! If not, I can suggest some conversation topics. Would you like to hear some suggestions? Just let me know! 😊criptions? 🤔 Would you like me to suggest some conversation topics? I'M HERE TO HELP! 😊 Just let me know! 🤗)ns topics? 😁)ns? Boise"

Which does not respect the regex, while:
client.text_generation("ok", grammar={"type": "regex", "value": ".+?\nCode"})
Gives a correct output:
'Question and the answer is: $\\boxed{0}$.\nCode'