[TLM] Update max_tokens for TLM to match provider limits

.github/workflows/ci-rerun-failed-test-tlm.yml

@@ -47,6 +47,7 @@ jobs:
         run: |
           pytest tests/tlm/test_properties.py --last-failed --verbose
       - name: Cache pytest results
+        if: always() && github.event.comment.body == '/rerun-failed-test-tlm' # Ensure this runs even if tests above fail
         uses: actions/cache@v2
         with:
           path: .pytest_cache

.github/workflows/ci-test-tlm.yml

@@ -88,6 +88,7 @@ jobs:
         run: |
           pytest -n auto tests/tlm/test_properties.py --verbose
       - name: Cache pytest results
+        if: always() && github.event.comment.body == '/test-tlm' # Ensure this runs even if tests above fail
         uses: actions/cache@v2
         with:
           path: .pytest_cache

cleanlab_studio/internal/constants.py

@@ -14,7 +14,7 @@
 ]
 _TLM_DEFAULT_MODEL: str = "gpt-4o-mini"
 _TLM_MAX_RETRIES: int = 3  # TODO: finalize this number
-TLM_MAX_TOKEN_RANGE: Tuple[int, int] = (64, 512)  # (min, max)
+TLM_MAX_TOKEN_RANGE: Tuple[int, int] = (64, 4096)  # (min, max)
 TLM_NUM_CANDIDATE_RESPONSES_RANGE: Tuple[int, int] = (1, 20)  # (min, max)
 TLM_NUM_CONSISTENCY_SAMPLES_RANGE: Tuple[int, int] = (0, 20)  # (min, max)
 TLM_VALID_LOG_OPTIONS: Set[str] = {"perplexity", "explanation"}