Adding support for O1

[ivanleomk]

This adds a new mode which should support the o1 model family using the openai client sdk. We'll need to upgrade the supported version to support the new max_completion_tokens parameter though. This allows us to do something like what we see below.

from instructor import from_openai, Mode
from openai import OpenAI
from pydantic import BaseModel

client = from_openai(OpenAI(), mode=Mode.JSON_O1)


class UserDetails(BaseModel):
    name: str
    age: int


resp = client.chat.completions.create(
    model="o1-mini",
    messages=[{"role": "user", "content": "Jason is 10"}],
    response_model=UserDetails,
    max_completion_tokens=300, # This has to have a minimum of 200+ because of reasoning tokens
)

print(resp)

---

🚀	This description was created by Ellipsis for commit f35b04c

feat: add support for JSON_O1 mode with OpenAI SDK

Summary:

Adds support for JSON_O1 mode using the OpenAI client SDK, requiring an updated openai dependency.

Key points:

• Behavior:
  • Adds support for JSON_O1 mode in from_openai() in client.py.
  • Updates handle_response_model() in process_response.py to handle JSON_O1 mode, ensuring no system messages are included.
  • Updates from_response() in function_calls.py to parse JSON_O1 mode using parse_json().
• Dependencies:
  • Updates openai dependency to ^1.45.0 in pyproject.toml to support max_completion_tokens.
• Enums:
  • Adds JSON_O1 to Mode enum in mode.py.

---

Generated with ❤️ by ellipsis.dev

[ellipsis-dev]

👍 Looks good to me! Reviewed everything up to f35b04c in 25 seconds

More details

• Looked at 89 lines of code in 5 files
• Skipped 0 files when reviewing.
• Skipped posting 2 drafted comments based on config settings.

1. instructor/client.py:426

• Draft comment:
  Consider adding specific handling for Mode.JSON_O1 in the from_openai function if it requires special processing, similar to other modes like MD_JSON or TOOLS_STRICT.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The PR introduces a new mode JSON_O1 but does not update the from_openai function to handle this mode specifically. This could lead to unexpected behavior if the mode requires special handling.

2. instructor/process_response.py:263

• Draft comment:
  Ensure that Mode.JSON_O1 is handled correctly in handle_response_model. The current implementation seems to append a message to new_kwargs["messages"], but verify if additional handling is needed.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The PR adds a new mode JSON_O1 but does not update the handle_response_model function to handle this mode specifically. This could lead to unexpected behavior if the mode requires special handling.

Workflow ID: wflow_YHsDjroDOx3yW1gZ

---

You can customize Ellipsis with 👍 / 👎 feedback, review rules, user-specific overrides, quiet mode, and more.

[cloudflare-workers-and-pages]

Deploying instructor with    Cloudflare Pages

Latest commit:	ee1c5af
Status:	✅  Deploy successful!
Preview URL:	https://e0dd1e53.instructor.pages.dev
Branch Preview URL:	https://support-o1.instructor.pages.dev

View logs

[ellipsis-dev]

Skipped PR review on ee1c5af because no changed files had a supported extension. If you think this was in error, please contact us and we'll fix it right away.

---

Generated with ❤️ by ellipsis.dev

[ivanleomk]

Important here to note that the current o1-preview doesn't support things like tool-calling, streaming and so this is really just supporting a simple json mode support