Bug Description

Every API request (/ask, /ask-llm, /ask-llm-prompted) produces
this warning in the server logs:

Both `max_new_tokens` (=1024) and `max_length`(=20) seem to 
have been set. `max_new_tokens` will take precedence. Please 
refer to the documentation for more information.

Environment

• transformers: 5.3.0
• torch: 2.10.0
• Python: 3.13
• OS: Windows 11

Root Cause Analysis

The current requirements.txt specifies:

transformers>=4.45.2

This allows transformers 5.x to be installed, which introduced
stricter generation config handling.

When transformers 5.x is installed, it detects TWO sources of
generation config conflicting:

Source 1 — model's downloaded generation_config.json:

{
  "top_k": 20,
  "temperature": 0.7,
  "top_p": 0.8,
  "repetition_penalty": 1.1
}

Source 2 — pipeline parameters in app/ai.py:

self.model = pipeline(
    "text-generation",
    model=model,
    max_new_tokens=1024,
    truncation=True,
    ...
)

transformers 4.x silently merged these two configs with no
warning. transformers 5.x is strict about this and raises
the warning on every single request.

Steps to Reproduce

1. Clone the repo
2. Install dependencies — transformers 5.3.0 gets installed
3. Run uvicorn main:app --host 0.0.0.0 --port 8000 --reload
4. Call POST /ask with any question
5. Warning appears twice in server logs for every request

Proposed Fix

Pin transformers below 5.0.0 in requirements.txt until the
codebase is updated to handle transformers 5.x config behavior:

# Before
transformers>=4.45.2

# After  
transformers>=4.45.2,<5.0.0

This is a one-line change that prevents other contributors from
hitting the same issue while a proper transformers 5.x
compatibility update is planned.

References

• HuggingFace transformers 5.x migration guide
• app/ai.py pipeline definitions (lines 59-80)
• requirements.txt line 1