sunholo 0.77.2__tar.gz → 0.77.3__tar.gz

{sunholo-0.77.2 → sunholo-0.77.3}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.77.2
+Version: 0.77.3
 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/v0.77.2.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.77.3.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.77.2 → sunholo-0.77.3}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.77.2'
+version = '0.77.3'
 
 setup(
     name='sunholo',

{sunholo-0.77.2 → sunholo-0.77.3}/sunholo/agents/flask/qna_routes.py

@@ -522,9 +522,11 @@ def prep_vac(request, vector_name):
     config, _ = load_config("config/llm_config.yaml")
     vac_configs = config.get("vac")
     if vac_configs:
-        vac_config = vac_configs[vector_name]
+        vac_config = vac_configs.get(vector_name)
+        if not vac_config:
+            log.warning("Not a local configured VAC, may be a remote config not synced yet")
 
-    if trace:
+    if trace and vac_config:
         trace.update(input=data, metadata=vac_config)
 
     user_input = data.pop('user_input').strip()

{sunholo-0.77.2 → sunholo-0.77.3}/sunholo/auth/run.py

@@ -2,22 +2,23 @@
 import inspect
 
 from typing import Dict, Optional
-from ..utils.config import load_config_key, load_config
+from ..utils.config import load_config
+from ..utils import ConfigManager
 from ..utils.gcp import is_running_on_cloudrun
 from ..utils.api_key import has_multivac_api_key, get_multivac_api_key
 from ..logging import log
 from ..agents.route import route_vac
 from .gcloud import get_local_gcloud_token
 
-def get_run_url(vector_name=None):
+def get_run_url(config):
 
-    if not vector_name:
+    if not config:
         raise ValueError('Vector name was not specified')
     
-    cloud_urls = route_vac(vector_name)
+    cloud_urls = route_vac(config=config)
     
     cloud_urls, _ = load_config('config/cloud_run_urls.json')
-    agent = load_config_key("agent", vector_name=vector_name, kind="vacConfig")
+    agent = config.vacConfig("agent")
 
     try:
         log.info(f'Looking up URL for {agent}')
@@ -44,10 +45,13 @@ def get_cloud_run_token(vector_name):
         
         return {"x-api-key": get_multivac_api_key()}
 
-    if is_running_on_cloudrun():
-        run_url = get_run_url(vector_name)
-    else:
-        run_url = "http://127.0.0.1:8080"
+    #if is_running_on_cloudrun():
+    #    run_url = get_run_url(vector_name)
+    #else:
+    #    run_url = "http://127.0.0.1:8080"
+    
+    config = ConfigManager(vector_name)
+    run_url = get_run_url(config)
 
     # Append ID Token to make authenticated requests to Cloud Run services
     frame = inspect.currentframe()

sunholo-0.77.3/sunholo/invoke/__init__.py

@@ -0,0 +1,2 @@
+from .invoke_vac_utils import invoke_vac
+from .direct_vac_func import direct_vac, direct_vac_stream

sunholo-0.77.3/sunholo/invoke/direct_vac_func.py

@@ -0,0 +1,129 @@
+from ..logging import log
+from ..agents import send_to_qa
+from ..qna.parsers import parse_output
+from ..streaming import generate_proxy_stream
+from ..utils import ConfigManager
+from ..utils.api_key import has_multivac_api_key
+
+def direct_vac(vac_input: dict, vac_name: str, chat_history=[]):
+    """
+    This lets VACs call other VAC Q&A endpoints within their code
+    """
+
+    log.info(f"Invoking VAC Q&A endpoints for {vac_name}")
+
+    if 'user_input' not in vac_input:
+        raise ValueError('vac_input must contain at least "user_input" key - got {vac_input}')
+
+    user_id = vac_input.get('user_id')
+    session_id = vac_input.get('session_id')
+    image_uri = vac_input.get('image_url') or vac_input.get('image_uri')
+
+    global_config = ConfigManager('global')
+    config = ConfigManager(vac_name)
+
+    agent_name = config.vacConfig('agent')
+    agent_url = config.vacConfig("agent_url")
+
+    if agent_url:
+        log.info("Found agent_url within vacConfig: {agent_url}")
+    # via public cloud endpoints - assumes no gcloud auth
+    override_endpoint = None
+    if has_multivac_api_key():
+        print("Found MULTIVAC_API_KEY")
+        gcp_config = global_config.vacConfig("gcp_config")
+        endpoints_base_url = gcp_config.get("endpoints_base_url")
+        if not endpoints_base_url:
+            raise ValueError("MULTIVAC_API_KEY env var is set but no config.gcp_config.endpoints_base_url can be found")
+
+        override_endpoint = f"{endpoints_base_url}/v1/{agent_name}"
+
+    override_endpoint = agent_url or override_endpoint
+
+    print(f"Using {override_endpoint=}")
+    log.warning(f'Batch invoke_vac_qa with {vac_input=}')
+    vac_response = send_to_qa(
+        vac_input["user_input"],
+        vector_name=vac_name,
+        chat_history=chat_history,
+        message_author=user_id,
+        #TODO: populate these
+        image_url=image_uri,
+        source_filters=None,
+        search_kwargs=None,
+        private_docs=None,
+        whole_document=False,
+        source_filters_and_or=False,
+        # system kwargs
+        configurable={
+            "vector_name": vac_name,
+        },
+        user_id=user_id,
+        session_id=session_id, 
+        message_source="sunholo.invoke_vac_qa.invoke",
+        override_endpoint=override_endpoint,
+        stream=False)
+        
+    # ensures {'answer': answer}
+    answer = parse_output(vac_response)
+    chat_history.append({"name": "Human", "content": vac_input})
+    chat_history.append({"name": "AI", "content": answer})
+    answer["chat_history"] = chat_history
+    
+    return answer
+
+def direct_vac_stream(vac_input: dict, vac_name: str, chat_history=[]):
+
+    if 'user_input' not in vac_input:
+        raise ValueError('vac_input must contain at least "user_input" key - got {vac_input}')
+
+    user_id = vac_input.get('user_id')
+    session_id = vac_input.get('session_id')
+    image_uri = vac_input.get('image_url') or vac_input.get('image_uri')
+    
+    log.info(f"Streaming invoke_vac_qa with {vac_input=}")
+    def stream_response():
+        generate = generate_proxy_stream(
+                send_to_qa,
+                vac_input["user_input"],
+                vector_name=vac_name,
+                chat_history=chat_history,
+                generate_f_output=lambda x: x,  # Replace with actual processing function
+                stream_wait_time=0.5,
+                stream_timeout=120,
+                message_author=user_id,
+                #TODO: populate these
+                image_url=image_uri,
+                source_filters=None,
+                search_kwargs=None,
+                private_docs=None,
+                whole_document=False,
+                source_filters_and_or=False,
+                # system kwargs
+                configurable={
+                    "vector_name": vac_name,
+                },
+                user_id=user_id,
+                session_id=session_id, 
+                message_source="sunholo.invoke_vac_qa.stream"
+        )
+        for part in generate():
+            yield part
+
+    answer = ""
+
+    for token in stream_response():
+        if isinstance(token, bytes):
+            token = token.decode('utf-8')
+            yield token
+        if isinstance(token, dict):
+            # ?
+            pass
+        elif isinstance(token, str):
+            answer += token
+
+    if answer:
+        chat_history.append({"name": "Human", "content": vac_input})
+        chat_history.append({"name": "AI", "content": answer})
+
+    return chat_history

sunholo-0.77.3/sunholo/invoke/invoke_vac_utils.py

@@ -0,0 +1,59 @@
+import json
+import requests
+
+from pathlib import Path
+
+from ..logging import log
+
+def invoke_vac(service_url, data, vector_name=None, metadata=None, is_file=False):
+    """
+    This lets a VAC be invoked by directly calling its URL, used for file uploads
+    """
+    try:
+        if is_file:
+            log.info("Uploading file...")
+            # Handle file upload
+            if not isinstance(data, Path) or not data.is_file():
+                raise ValueError("For file uploads, 'data' must be a Path object pointing to a valid file.")
+            
+            files = {
+                'file': (data.name, open(data, 'rb')),
+            }
+            form_data = {
+                'vector_name': vector_name,
+                'metadata': json.dumps(metadata) if metadata else '',
+            }
+
+            response = requests.post(service_url, files=files, data=form_data)
+        else:
+            log.info("Uploading JSON...")
+            try:
+                if isinstance(data, dict):
+                    json_data = data
+                else:
+                    json_data = json.loads(data)
+            except json.JSONDecodeError as err:
+                log.error(f"[bold red]ERROR: invalid JSON: {str(err)} [/bold red]")
+                raise err
+            except Exception as err:
+                log.error(f"[bold red]ERROR: could not parse JSON: {str(err)} [/bold red]")
+                raise err
+
+            log.debug(f"Sending data: {data} or json_data: {json.dumps(json_data)}")
+            # Handle JSON data
+            headers = {"Content-Type": "application/json"}
+            response = requests.post(service_url, headers=headers, data=json.dumps(json_data))
+
+        response.raise_for_status()
+
+        the_data = response.json()
+        log.info(the_data)
+
+        return the_data
+    
+    except requests.exceptions.RequestException as e:
+        log.error(f"[bold red]ERROR: Failed to invoke VAC: {e}[/bold red]")
+        raise e
+    except Exception as e:
+        log.error(f"[bold red]ERROR: An unexpected error occurred: {e}[/bold red]")
+        raise e

{sunholo-0.77.2 → sunholo-0.77.3}/sunholo.egg-info/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.77.2
+Version: 0.77.3
 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/v0.77.2.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.77.3.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.77.2 → sunholo-0.77.3}/sunholo.egg-info/SOURCES.txt

@@ -90,6 +90,7 @@ sunholo/gcs/download_folder.py
 sunholo/gcs/download_url.py
 sunholo/gcs/metadata.py
 sunholo/invoke/__init__.py
+sunholo/invoke/direct_vac_func.py
 sunholo/invoke/invoke_vac_utils.py
 sunholo/langfuse/__init__.py
 sunholo/langfuse/callback.py

sunholo-0.77.2/sunholo/invoke/__init__.py

@@ -1 +0,0 @@
-from .invoke_vac_utils import invoke_vac, invoke_vac_qa

sunholo-0.77.2/sunholo/invoke/invoke_vac_utils.py

@@ -1,152 +0,0 @@
-import json
-import requests
-
-from pathlib import Path
-
-from ..logging import log
-from ..agents import send_to_qa
-from ..qna.parsers import parse_output
-from ..streaming import generate_proxy_stream
-
-def invoke_vac_qa(vac_input: dict, vac_name: str, chat_history=[], stream=False):
-    """
-    This lets VACs call other VAC Q&A endpoints within their code
-    """
-
-    if 'user_input' not in vac_input:
-        raise ValueError('vac_input must contain at least "user_input" key - got {vac_input}')
-
-    user_id = vac_input.get('user_id')
-    session_id = vac_input.get('session_id')
-    image_uri = vac_input.get('image_url') or vac_input.get('image_uri')
-
-    if not stream:
-        log.warning(f'Batch invoke_vac_qa with {vac_input=}')
-        vac_response = send_to_qa(
-            vac_input["user_input"],
-            vector_name=vac_name,
-            chat_history=chat_history,
-            message_author=user_id,
-            #TODO: populate these
-            image_url=image_uri,
-            source_filters=None,
-            search_kwargs=None,
-            private_docs=None,
-            whole_document=False,
-            source_filters_and_or=False,
-            # system kwargs
-            configurable={
-                "vector_name": vac_name,
-            },
-            user_id=user_id,
-            session_id=session_id, 
-            message_source="sunholo.invoke_vac_qa.invoke",
-            stream=False)
-        
-        # ensures {'answer': answer}
-        answer = parse_output(vac_response)
-        chat_history.append({"name": "Human", "content": vac_input})
-        chat_history.append({"name": "AI", "content": answer})
-        answer["chat_history"] = chat_history
-        
-        return answer
-    
-    log.info(f"Streaming invoke_vac_qa with {vac_input=}")
-    def stream_response():
-        generate = generate_proxy_stream(
-                send_to_qa,
-                vac_input["user_input"],
-                vector_name=vac_name,
-                chat_history=chat_history,
-                generate_f_output=lambda x: x,  # Replace with actual processing function
-                stream_wait_time=0.5,
-                stream_timeout=120,
-                message_author=user_id,
-                #TODO: populate these
-                image_url=image_uri,
-                source_filters=None,
-                search_kwargs=None,
-                private_docs=None,
-                whole_document=False,
-                source_filters_and_or=False,
-                # system kwargs
-                configurable={
-                    "vector_name": vac_name,
-                },
-                user_id=user_id,
-                session_id=session_id, 
-                message_source="sunholo.invoke_vac_qa.stream"
-        )
-        for part in generate():
-            yield part
-
-    answer = ""
-
-    for token in stream_response():
-        if isinstance(token, bytes):
-            token = token.decode('utf-8')
-            yield token
-        if isinstance(token, dict):
-            # ?
-            pass
-        elif isinstance(token, str):
-            answer += token
-
-    if answer:
-        chat_history.append({"name": "Human", "content": vac_input})
-        chat_history.append({"name": "AI", "content": answer})
-
-    return chat_history
-
-def invoke_vac(service_url, data, vector_name=None, metadata=None, is_file=False):
-    """
-    This lets a VAC be invoked by directly calling its URL, used for file uploads
-    """
-    try:
-        if is_file:
-            log.info("Uploading file...")
-            # Handle file upload
-            if not isinstance(data, Path) or not data.is_file():
-                raise ValueError("For file uploads, 'data' must be a Path object pointing to a valid file.")
-            
-            files = {
-                'file': (data.name, open(data, 'rb')),
-            }
-            form_data = {
-                'vector_name': vector_name,
-                'metadata': json.dumps(metadata) if metadata else '',
-            }
-
-            response = requests.post(service_url, files=files, data=form_data)
-        else:
-            log.info("Uploading JSON...")
-            try:
-                if isinstance(data, dict):
-                    json_data = data
-                else:
-                    json_data = json.loads(data)
-            except json.JSONDecodeError as err:
-                log.error(f"[bold red]ERROR: invalid JSON: {str(err)} [/bold red]")
-                raise err
-            except Exception as err:
-                log.error(f"[bold red]ERROR: could not parse JSON: {str(err)} [/bold red]")
-                raise err
-
-            log.debug(f"Sending data: {data} or json_data: {json.dumps(json_data)}")
-            # Handle JSON data
-            headers = {"Content-Type": "application/json"}
-            response = requests.post(service_url, headers=headers, data=json.dumps(json_data))
-
-        response.raise_for_status()
-
-        the_data = response.json()
-        log.info(the_data)
-
-        return the_data
-    
-    except requests.exceptions.RequestException as e:
-        log.error(f"[bold red]ERROR: Failed to invoke VAC: {e}[/bold red]")
-        raise e
-    except Exception as e:
-        log.error(f"[bold red]ERROR: An unexpected error occurred: {e}[/bold red]")
-        raise e