PyPI - hana-ai - Versions diffs - 1.0.25070700.post2__tar.gz → 1.0.25090400__tar.gz - Mend

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hana_ai
-Version: 1.0.25070700.post2
+Version: 1.0.25090400
 Summary: Generative AI Toolkit for SAP HANA Cloud
 Author: SAP
 License: Apache-2.0

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "hana_ai"
-version = "1.0.25070700.post2"
+version = "1.0.25090400"
 authors = [
     {name = "SAP"}
 ]

@@ -2,4 +2,4 @@
 hana_ai is a Python package for AI/ML related utilities.
 """
-__version__ = "1.0.25070700.post2"
+__version__ = "1.0.25090400"

@@ -52,16 +52,9 @@ class GetCodeTemplateFromVectorDB(BaseTool):
     def _run(
         self,
-        **kwargs
+        query
     ) -> str:
         """Use the tool."""
-        if "kwargs" in kwargs:
-            kwargs = kwargs["kwargs"]
-        query = kwargs.get("query", None)
-        if query is None:
-            return "Query is required"
         if self.vectordb is None:
             raise ValueError("No vector database set.")
         model = self.vectordb
@@ -70,7 +63,7 @@ class GetCodeTemplateFromVectorDB(BaseTool):
         return result
     async def _arun(
-        self, **kwargs
+        self, query
     ) -> str:
         """Use the tool asynchronously."""
-        return self._run(**kwargs)
+        return self._run(query)

@@ -0,0 +1,106 @@
+"""
+Cross Encoder inside HANA
+    * :class:`PALCrossEncoder`
+"""
+import uuid
+import logging
+from hdbcli import dbapi
+import pandas as pd
+from hana_ml.dataframe import create_dataframe_from_pandas, DataFrame
+from hana_ml.algorithms.pal.pal_base import (
+    PALBase,
+    ParameterTable,
+    try_drop
+)
+# pylint: disable=line-too-long, super-with-arguments, too-many-arguments, too-many-positional-arguments, consider-using-f-string
+logger = logging.getLogger(__name__)
+class PALCrossEncoder(PALBase):
+    """
+    PAL embeddings base class.
+    """
+    def __init__(self, connection_context):
+        super(PALCrossEncoder, self).__init__()
+        self.connection_context = connection_context
+        self.stats_ = None
+    def _predict(self, data, thread_number=None, batch_size=None, max_token_num=None, model_version=None):
+        """
+        Reranking with Cross Encoder
+        """
+        unique_id = str(uuid.uuid1()).replace('-', '_').upper()
+        result_tbl = '#PAL_CROSSENCODER_RESULT_TBL_{}_{}'.format(0, unique_id)
+        stats_tbl = '#PAL_CROSSENCODER_STATS_TBL_{}_{}'.format(0, unique_id)
+        outputs = [result_tbl, stats_tbl]
+        param_rows = [("BATCH_SIZE", batch_size, None, None),
+                      ("MODEL_VERSION", None, None, model_version),
+                      ("THREAD_NUMBER", thread_number, None, None),
+                      ("MAX_TOKEN_NUM", max_token_num, None, None)]
+        try:
+            self._call_pal_auto(self.connection_context,
+                                'PAL_CROSSENCODER',
+                                data,
+                                ParameterTable().with_data(param_rows),
+                                *outputs)
+        except dbapi.Error as db_err:
+            msg = str(self.connection_context.hana_version())
+            logger.exception("HANA version: %s. %s", msg, str(db_err))
+            try_drop(self.connection_context, outputs)
+            raise
+        except Exception as db_err:
+            logger.exception(str(db_err))
+            try_drop(self.connection_context, outputs)
+            raise
+        self.stats_ = self.connection_context.table(stats_tbl)
+        return self.connection_context.table(result_tbl)
+    def predict(self, data, thread_number=None, batch_size=None, max_token_num=None, model_version=None, return_table=False):
+        """
+        Reranking with Cross Encoder
+        """
+        if isinstance(data, list):
+            # data is list of (query, content)
+            # have to upload data to hana table with columns id, query, content
+            df = pd.DataFrame(data, columns=['QUERY', 'CONTENT'])
+            # add id column to df
+            df.insert(0, 'ID', range(len(df)))
+            # create temporary table name with uuid
+            temporary_table = "#PAL_CROSSENCODER_INPUT_TBL_" + str(uuid.uuid4()).replace("-", "_")
+            data_ = create_dataframe_from_pandas(self.connection_context, df, table_name=temporary_table, disable_progressbar=True)
+        elif isinstance(data, DataFrame):
+            data_ = data
+        else:
+            raise ValueError("data should be list of (query, content) or hana_ml.dataframe.DataFrame with (id, query, content) columns")
+        result = self._predict(data_,
+                               thread_number=thread_number,
+                               batch_size=batch_size,
+                               max_token_num=max_token_num,
+                               model_version=model_version)
+        if return_table:
+            return result
+        # fetch result to pandas dataframe by collect with first column as id and second column as score
+        result_df = result.collect()
+        try_drop(self.connection_context, temporary_table)
+        # sort result_df by id
+        result_df = result_df.sort_values(by=result_df.columns[0])
+        # return score column as ndarray
+        return result_df[result_df.columns[1]].to_numpy()
+# if __name__ == "__main__":
+#     from hana_ml import ConnectionContext
+#     address = "8f4b6bad-5c7e-4673-a41c-d59d78b4aa4f.hana.aws.hcd-us10.hanacloud.ondemand.com"
+#     port = 443
+#     user = "PALDEVUSER"
+#     password = "Abcd1234"
+#     cc = ConnectionContext(address=address, port=port, user=user, password=password)
+#     ce = PALCrossEncoder(cc)
+#     data = [("What is HANA?", "HANA is a database"), ("What is Python?", "Python is a programming language")]
+#     result = ce.predict(data, return_table=True)
+#     if isinstance(result, DataFrame):
+#         print(result.collect())
+#     else:
+#         print(result)
+#     print(ce.stats_.collect())

@@ -43,6 +43,7 @@ src/hana_ai/vectorstore/code_templates.py
 src/hana_ai/vectorstore/corrective_retriever.py
 src/hana_ai/vectorstore/embedding_service.py
 src/hana_ai/vectorstore/hana_vector_engine.py
+src/hana_ai/vectorstore/pal_cross_encoder.py
 src/hana_ai/vectorstore/union_vector_stores.py
 src/hana_ai/vectorstore/knowledge_base/__init__.py
 src/hana_ai/vectorstore/knowledge_base/python_knowledge/AMDPDeployer.txt

hana-ai 1.0.25070700.post2__tar.gz → 1.0.25090400__tar.gz

hana-ai 1.0.25070700.post2tar.gz → 1.0.25090400tar.gz