sunholo 0.62.13__tar.gz → 0.62.15__tar.gz

{sunholo-0.62.13 → sunholo-0.62.15}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.62.13
+Version: 0.62.15
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.13.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.15.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.62.13 → sunholo-0.62.15}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.62.13'
+version = '0.62.15'
 
 setup(
     name='sunholo',

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo/agents/flask/qna_routes.py

@@ -45,7 +45,6 @@ def register_qna_routes(app, stream_interpreter, vac_interpreter):
         return jsonify({"status": "healthy"})
 
     @app.route('/vac/streaming/<vector_name>', methods=['POST'])
-    @observe()
     def stream_qa(vector_name):
         observed_stream_interpreter = observe()(stream_interpreter)
         prep = prep_vac(request, vector_name)
@@ -111,8 +110,8 @@ def register_qna_routes(app, stream_interpreter, vac_interpreter):
         return response
 
     @app.route('/vac/<vector_name>', methods=['POST'])
-    @observe()
     def process_qna(vector_name):
+        observed_vac_interpreter = observe()(vac_interpreter)
         prep = prep_vac(request, vector_name)
         log.debug(f"Processing prep: {prep}")
         trace = prep["trace"]
@@ -132,13 +131,13 @@ def register_qna_routes(app, stream_interpreter, vac_interpreter):
                     input = all_input,
                     model=vac_config.get("model") or vac_config.get("llm")
                 )
-            bot_output = vac_interpreter(
+            bot_output = observed_vac_interpreter(
                 question=all_input["user_input"],
                 vector_name=vector_name,
                 chat_history=all_input["chat_history"],
                 **all_input["kwargs"]
             )
-            if generation:
+            if span:
                 generation.end(output=bot_output)
             # {"answer": "The answer", "source_documents": [{"page_content": "The page content", "metadata": "The metadata"}]}
             bot_output = parse_output(bot_output)

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo/cli/run_proxy.py

@@ -277,7 +277,7 @@ def setup_proxy_subparser(subparsers):
     start_parser.add_argument('--port', type=int, help='Port to run the proxy on. Auto-assigns if not provided.')
     start_parser.add_argument('--local', action='store_true', help='Run the service locally instead of proxying to Cloud Run.')
     start_parser.add_argument('--app-type', choices=['flask', 'fastapi'], help='If local, type of the local app (flask or fastapi).')
-    start_parser.add_argument('--app-folder', help='If local, folder containing the local app.py')
+    start_parser.add_argument('--app-folder', default=".", help='If local, folder containing the local app.py')
     start_parser.add_argument('--log-file', action='store_true', help='Whether to create a file containing proxy logs.')
     start_parser.set_defaults(func=lambda args: start_proxy(args.service_name, 
                                                             args.region, 

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo/components/retriever.py

@@ -29,7 +29,7 @@ from langchain.retrievers import ContextualCompressionRetriever
 def load_memories(vector_name):
     memories = load_config_key("memory", vector_name, kind="vacConfig")
     log.info(f"Found memory settings for {vector_name}: {memories}")
-    if len(memories) == 0:
+    if not memories or len(memories) == 0:
         log.info(f"No memory settings found for {vector_name}")
         return None
 
@@ -67,7 +67,7 @@ def pick_retriever(vector_name, embeddings=None):
             
             #TODO: more memory stores here
 
-    if len(retriever_list) == 0:
+    if not retriever_list or len(retriever_list) == 0:
         log.info(f"No retrievers were created for {memories}")
         return None
     

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo/qna/parsers.py

@@ -36,7 +36,7 @@ def parse_output(bot_output):
 
         return bot_output
     
-    elif isinstance(bot_output, dict) and 'metadata' in bot_output and isinstance(bot_output.get('metadata')) and 'source_documents' in bot_output.get('metadata'):
+    elif isinstance(bot_output, dict) and 'metadata' in bot_output and isinstance(bot_output.get('metadata'), dict) and 'source_documents' in bot_output.get('metadata'):
         metadata = bot_output.get('metadata')
         bot_output['source_documents'] = [document_to_dict(doc) for doc in metadata['source_documents']]
         if not bot_output.get("answer") or bot_output.get("answer") == "":

sunholo-0.62.15/sunholo/vertex/__init__.py

@@ -0,0 +1,3 @@
+from .init import init_vertex
+from .memory_tools import get_vertex_memories, print_grounding_response
+from .safety import vertex_safety

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo/vertex/memory_tools.py

@@ -1,6 +1,6 @@
 try:
     from vertexai.preview import rag
-    from vertexai.preview.generative_models import Tool, grounding
+    from vertexai.preview.generative_models import Tool, grounding, GenerationResponse
 except ImportError:
     rag = None
 
@@ -49,6 +49,10 @@ def get_vertex_memories(vector_name):
 
     memories = load_memories(vector_name)
     tools = []
+
+    if not memories:
+        return tools
+    
     for memory in memories:
         for key, value in memory.items():  # Now iterate over the dictionary
             log.info(f"Found memory {key}")
@@ -89,3 +93,54 @@ def get_vertex_memories(vector_name):
         log.warning("No llamaindex Vertex corpus configurations could be found")
     
     return tools
+
+def print_grounding_response(response):
+    """Prints Gemini response with grounding citations."""
+    grounding_metadata = response.candidates[0].grounding_metadata
+
+    # Citation indices are in byte units
+    ENCODING = "utf-8"
+    text_bytes = response.text.encode(ENCODING)
+
+    prev_index = 0
+    markdown_text = ""
+
+    sources: dict[str, str] = {}
+    footnote = 1
+    for attribution in grounding_metadata.grounding_attributions:
+        context = attribution.web or attribution.retrieved_context
+        if not context:
+            log.info(f"Skipping Grounding Attribution {attribution}")
+            continue
+
+        title = context.title
+        uri = context.uri
+        end_index = int(attribution.segment.end_index)
+
+        if uri not in sources:
+            sources[uri] = {"title": title, "footnote": footnote}
+            footnote += 1
+
+        text_segment = text_bytes[prev_index:end_index].decode(ENCODING)
+        markdown_text += f"{text_segment} [[{sources[uri]['footnote']}]]({uri})"
+        prev_index = end_index
+
+    if prev_index < len(text_bytes):
+        markdown_text += str(text_bytes[prev_index:], encoding=ENCODING)
+
+    markdown_text += "\n## Grounding Sources\n"
+
+    if grounding_metadata.web_search_queries:
+        markdown_text += (
+            f"\n**Web Search Queries:** {grounding_metadata.web_search_queries}\n"
+        )
+    elif grounding_metadata.retrieval_queries:
+        markdown_text += (
+            f"\n**Retrieval Queries:** {grounding_metadata.retrieval_queries}\n"
+        )
+
+    for uri, source in sources.items():
+        markdown_text += f"{source['footnote']}. [{source['title']}]({uri})\n"
+    
+    log.info(markdown_text)
+    return markdown_text

sunholo-0.62.15/sunholo/vertex/safety.py

@@ -0,0 +1,35 @@
+try:
+    from vertexai.generative_models import (
+        HarmCategory,
+        HarmBlockThreshold,
+    )
+except ImportError:
+    pass
+
+def vertex_safety(threshold: str = "BLOCK_ONLY_HIGH"):
+    """
+    BLOCK_ONLY_HIGH - block when high probability of unsafe content is detected
+    BLOCK_MEDIUM_AND_ABOVE - block when medium or high probability of content is detected
+    BLOCK_LOW_AND_ABOVE - block when low, medium, or high probability of unsafe content is detected
+    BLOCK_NONE - no block, but need to be on an allow list to use
+    """
+
+    if threshold == 'BLOCK_ONLY_HIGH':
+        thresh = HarmBlockThreshold.BLOCK_ONLY_HIGH
+    elif threshold == 'BLOCK_MEDIUM_AND_ABOVE':
+        thresh = HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE
+    elif threshold == 'BLOCK_LOW_AND_ABOVE':
+        thresh = HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
+    elif threshold == 'BLOCK_NONE':
+        thresh = HarmBlockThreshold.BLOCK_NONE
+    else:
+        raise ValueError("Invalid threshold")
+
+    safety_settings = {
+        HarmCategory.HARM_CATEGORY_HARASSMENT: thresh,
+        HarmCategory.HARM_CATEGORY_HATE_SPEECH: thresh,
+        HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: thresh,
+        HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: thresh,
+    }
+
+    return safety_settings

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo.egg-info/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.62.13
+Version: 0.62.15
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.13.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.15.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.62.13 → sunholo-0.62.15}/sunholo.egg-info/SOURCES.txt

@@ -111,6 +111,7 @@ sunholo/utils/user_ids.py
 sunholo/vertex/__init__.py
 sunholo/vertex/init.py
 sunholo/vertex/memory_tools.py
+sunholo/vertex/safety.py
 tests/test_chat_history.py
 tests/test_config.py
 tests/test_dispatch_to_qa.py

{sunholo-0.62.13 → sunholo-0.62.15}/tests/test_dispatch_to_qa.py

@@ -4,7 +4,11 @@ from sunholo.agents.dispatch_to_qa import prep_request_payload
 
 @pytest.mark.parametrize("user_input, chat_history, vector_name, stream, expected_endpoint, expected_payload", [
     ("What is AI?", [], "my_vector", False, "http://example.com/invoke", {"user_input": "What is AI?", "chat_history": [], "vector_name": "my_vector"}),
-    ("Tell me about ML", ["Previous chat message"], "another_vector", True, "http://example.com/stream", {"user_input": "Tell me about ML", "chat_history": ["Previous chat message"], "vector_name": "another_vector"})
+    ("Tell me about ML", ["Previous chat message"], "another_vector", True, "http://example.com/stream", {"user_input": "Tell me about ML", "chat_history": ["Previous chat message"], "vector_name": "another_vector"}),
+    # Additional test cases to cover all logic paths
+    ("", [], "empty_input_vector", False, "http://example.com/invoke", {"user_input": "", "chat_history": [], "vector_name": "empty_input_vector"}),
+    ("Valid input", ["Chat history present"], "history_vector", False, "http://example.com/invoke", {"user_input": "Valid input", "chat_history": ["Chat history present"], "vector_name": "history_vector"}),
+    ("Stream request", [], "stream_vector", True, "http://example.com/stream", {"user_input": "Stream request", "chat_history": [], "vector_name": "stream_vector"})
 ])
 @patch('sunholo.agents.dispatch_to_qa.load_config_key')
 @patch('sunholo.agents.dispatch_to_qa.route_endpoint')

sunholo-0.62.13/sunholo/vertex/__init__.py

@@ -1,2 +0,0 @@
-from .init import init_vertex
-from .memory_tools import get_vertex_memories