llm-join 0.2.4__tar.gz → 0.2.6__tar.gz

{llm_join-0.2.4 → llm_join-0.2.6}/PKG-INFO

@@ -1,12 +1,12 @@
 Metadata-Version: 2.4
 Name: llm-join
-Version: 0.2.4
+Version: 0.2.6
 Summary: Fuzzy join DataFrames using LLM scoring and embedding retrieval
 License: MIT
 Keywords: entity-resolution,fuzzy-join,llm,nlp,pandas
 Requires-Python: >=3.9
-Requires-Dist: faiss-cpu>=1.8
-Requires-Dist: numpy>=1.23
+Requires-Dist: faiss-cpu<1.14,>=1.8
+Requires-Dist: numpy<2,>=1.23
 Requires-Dist: pandas>=1.5
 Provides-Extra: sentence-transformers
 Requires-Dist: sentence-transformers>=2.2; extra == 'sentence-transformers'

{llm_join-0.2.4 → llm_join-0.2.6}/llm_join/__init__.py

@@ -5,8 +5,11 @@ except ImportError as e:
     if "numpy._core" in str(e) or "numpy.core" in str(e):
         raise ImportError(
             "llm-join: numpy/faiss version mismatch detected.\n\n"
+            "faiss-cpu>=1.14 requires numpy>=2, which breaks numpy 1.x environments.\n\n"
             "Fix with:\n"
-            "  pip install 'faiss-cpu>=1.8' --force-reinstall\n\n"
+            "  pip install \"faiss-cpu>=1.8,<1.14\" \"numpy>=1.23,<2\" --force-reinstall\n\n"
+            "On Databricks, run the above as %pip install ... then restart Python:\n"
+            "  dbutils.library.restartPython()\n\n"
             f"Original error: {e}"
         ) from None
     raise
@@ -14,4 +17,4 @@ except ImportError as e:
 from llm_join.join import fuzzy_join
 
 __all__ = ["fuzzy_join"]
-__version__ = "0.2.4"
+__version__ = "0.2.6"

{llm_join-0.2.4 → llm_join-0.2.6}/llm_join/join.py

@@ -106,6 +106,7 @@ def fuzzy_join(
                 reasoning="LLM failed — embed rank-0 fallback used",
                 embed_rank=0,
                 match_method="embed_fallback",
+                candidates=candidates,
             ))
         elif results:
             matches.extend(results)

{llm_join-0.2.4 → llm_join-0.2.6}/llm_join/merger.py

@@ -9,7 +9,8 @@ class MatchResult:
     score: float
     reasoning: str
     embed_rank: int = 0
-    match_method: str = "llm"  # "llm" or "embed_threshold"
+    match_method: str = "llm"  # "llm", "embed_threshold", or "embed_fallback"
+    candidates: list = None  # top-K candidates sent to LLM (None if LLM was not called)
 
 
 class Merger:
@@ -50,6 +51,7 @@ class Merger:
             "_llm_reasoning": [m.reasoning for m in matches],
             "_embed_rank": [m.embed_rank for m in matches],
             "_match_method": [m.match_method for m in matches],
+            "_llm_candidates": [m.candidates for m in matches],
         })
 
         df2_with_key = df2.merge(match_df, on=right_col, how="left")
@@ -57,7 +59,7 @@ class Merger:
 
         if not return_reasoning:
             result = result.drop(
-                columns=["_llm_score", "_llm_reasoning", "_embed_rank", "_match_method"],
+                columns=["_llm_score", "_llm_reasoning", "_embed_rank", "_match_method", "_llm_candidates"],
                 errors="ignore",
             )
 

{llm_join-0.2.4 → llm_join-0.2.6}/llm_join/scorer.py

@@ -154,6 +154,7 @@ class LLMScorer:
                 score=best_score,
                 reasoning=reasoning + tie_note,
                 embed_rank=idx,
+                candidates=candidates,
             )
             for idx, reasoning in tied
         ]

{llm_join-0.2.4 → llm_join-0.2.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "llm-join"
-version = "0.2.4"
+version = "0.2.6"
 description = "Fuzzy join DataFrames using LLM scoring and embedding retrieval"
 readme = "README.md"
 requires-python = ">=3.9"
@@ -12,8 +12,8 @@ license = { text = "MIT" }
 keywords = ["fuzzy-join", "llm", "entity-resolution", "pandas", "nlp"]
 dependencies = [
     "pandas>=1.5",
-    "numpy>=1.23",
-    "faiss-cpu>=1.8",
+    "numpy>=1.23,<2",
+    "faiss-cpu>=1.8,<1.14",
 ]
 
 [project.optional-dependencies]