mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-14 23:21:35 +00:00
b4a9ddb924
The transform now declares reasoning_effort_levels straight from the catalog's effort options instead of mapping them onto per-level support flags, which mis-advertised efforts these models do not take. max_tokens mirrors max_completion_tokens rather than context_length, supports_prompt_caching is derived from cache-read pricing, video input is read from input_modalities, and a failed Friendli fetch no longer breaks the weekly sync run. The committed friendliai entries are regenerated from the live catalog: stale gemma-4 token caps refreshed, the delisted K-EXAONE-236B-A23B entry dropped, and GLM-5.3 / GLM-5.3-Flash picked up with current billed pricing. |
||
|---|---|---|
| .. | ||
| _agent_shin_actions.py | ||
| agent_shin_shared.py | ||
| assert_ci_coverage.py | ||
| assert_workflow_dir_hygiene.py | ||
| auto_update_price_and_context_window_file.py | ||
| close_duplicate_issues.py | ||
| close_low_quality_prs.py | ||
| detect_changes.sh | ||
| e2e_egress_sentinel.py | ||
| e2e_fetch_fixture_bundle.sh | ||
| e2e_pack_fixture_bundle.sh | ||
| run_llm_translation_tests.py | ||
| scan_keywords.py | ||
| select_ui_test_scope.sh | ||
| triage-requirements.txt | ||
| triage_with_llm.py | ||
| uv_sync_with_retries.sh | ||