fix(llm): force temperature to 1.0 for gpt-5/o1/o3 models #141

[ahmedjawedaj]

• Added forced_temperature capability in capabilities.py
• Implemented override logic in cloud_provider.py for complete() and stream()
• Added unit tests verifying override behavior

Description

A clear and concise description of the changes.

Related Issues

• Closes #...
• Related to #...

Module(s) Affected

• agents
• api
• config
• core
• knowledge
• logging
• services
• tools
• utils
• web (Frontend)
• docs (Documentation)
• scripts
• tests
• Other: ...

Checklist

• I have read and followed the contribution guidelines.
• My code follows the project's coding standards.
• I have run pre-commit run --all-files and fixed any issues.
• I have added relevant tests for my changes.
• I have updated the documentation (if necessary).
• My changes do not introduce any new security vulnerabilities.

Additional Notes

Add any other context or screenshots about the pull request here.

[scrrlt]

requested changes:

• move all parameter sanitization logic to new file src/services/llm/parameters.py.
• centralize logic here to avoid cluttering cloud_provider.py.

src/services/llm/capabilities.py

• remove speculative entries (gpt-5, o3).
• add MODEL_CONSTRAINTS dict for strict rules (e.g. temperature_strategy).
• keep get_capability for feature flags.
• add get_model_constraint for enforcing limits.

3. src/services/llm/parameters.py (new)

• implement sanitize_model_params(binding, model, kwargs).
• check temperature_strategy:
• fixed: overwrite user input (e.g. o1 -> 1.0).
• drop: remove key if model rejects it.
• resolve max_tokens vs max_completion_tokens dynamically.
• import/define DEFAULT_MAX_TOKENS constant (no magic numbers).

4. src/services/llm/cloud_provider.py

• import sanitize_model_params.
• remove inline logic (the if forced_temp block).
• remove hardcoded 4096 fallback.
• delegate payload construction to parameters.py.

5. src/services/llm/config.py

ensure get_token_limit_kwargs correctly maps keys based on model version (legacy max_tokens vs new max_completion_tokens).

6. testing

• drop complex asyncmock setup for this feature.
• add tests/services/llm/test_parameters.py.
• write pure unit tests (table-driven) to verify logic in isolation.

in the midst of refactoring llm section so this would make my life much easier!! <3 added test is good practice :)

[pancacake]

Thanks for your contribution! Sadly there are some conflicts with our current repo. We've fixed it in the newest update, please refers to: 51902b1