sunholo 0.62.2__tar.gz → 0.62.4__tar.gz

{sunholo-0.62.2 → sunholo-0.62.4}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.62.2
+Version: 0.62.4
 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/v0.62.2.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.4.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.62.2 → sunholo-0.62.4}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.62.2'
+version = '0.62.4'
 
 setup(
     name='sunholo',

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/agents/dispatch_to_qa.py

@@ -59,7 +59,10 @@ def prep_request_payload(user_input, chat_history, vector_name, stream, **kwargs
     # {'stream': '', 'invoke': ''}
     endpoints = route_endpoint(vector_name, override_endpoint=override_endpoint)
 
-    qna_endpoint = endpoints["stream"] if stream else endpoints["invoke"]
+    if stream:
+        qna_endpoint = endpoints["stream"]
+    else: 
+        qna_endpoint = endpoints["invoke"]
 
     if agent == "langserve" or agent_type == "langserve":
         qna_data = prepare_request_data(user_input, endpoints["input_schema"], vector_name, **kwargs)
@@ -67,6 +70,7 @@ def prep_request_payload(user_input, chat_history, vector_name, stream, **kwargs
         # Base qna_data dictionary
         qna_data = {
             'user_input': user_input,
+            'vector_name': vector_name,
         }
         # Update qna_data with optional values from kwargs
         qna_data.update(kwargs)

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/agents/route.py

@@ -42,15 +42,8 @@ def route_endpoint(vector_name, override_endpoint=None):
         agent_type = load_config_key('agent', vector_name, kind="vacConfig")
 
     stem = route_qna(vector_name) if not override_endpoint else override_endpoint
-
-    agent_config, _ = load_config('config/agent_config.yaml')
-
-    # Select the appropriate configuration based on agent_type
-    if agent_type in agent_config:
-        endpoints_config = agent_config[agent_type]
-    else:
-        log.warning('Using default endpoints configuration')
-        endpoints_config = agent_config['default']
+    
+    endpoints_config = load_config_key(agent_type, vector_name, kind="agentConfig")
     
     log.info(f"endpoints_config: {endpoints_config}")
 

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/cli/chat_vac.py

@@ -1,7 +1,8 @@
 from ..agents import send_to_qa
-from ..streaming import generate_proxy_stream
+from ..streaming import generate_proxy_stream, can_agent_stream
 from ..utils.user_ids import generate_user_id
 from ..utils.config import load_config_key
+from ..logging import log
 
 from .run_proxy import clean_proxy_list, start_proxy, stop_proxy
 
@@ -33,12 +34,16 @@ def get_service_url(vac_name, project, region, no_config=False):
         port = proxies[agent_name]['port']
         url = f"http://127.0.0.1:{port}"
     else:
-        console.print(f"No proxy found running for service: [bold orange]'{agent_name}[/bold orange] required for [bold orange]{vac_name}[/bold orange] - attempting to connect")
-        url = start_proxy(agent_name, region, project)
+        if agent_name:
+            console.print(f"No proxy found running for service: [bold orange]'{agent_name}'[/bold orange] required for [bold orange]{vac_name}[/bold orange] - attempting to connect")
+            url = start_proxy(agent_name, region, project)
+        else:
+            console.print(f"No config for [bold orange]'{vac_name}'[/bold orange] - can't start proxy")
+            sys.exit(1)
 
     return url
 
-def stream_chat_session(service_url, service_name):
+def stream_chat_session(service_url, service_name, stream=True):
 
     user_id = generate_user_id()
     chat_history = []
@@ -48,16 +53,11 @@ def stream_chat_session(service_url, service_name):
         if user_input.lower() in ["exit", "quit"]:
             console.print("[bold red]Exiting chat session.[/bold red]")
             break
-
-        def stream_response():
-            generate = generate_proxy_stream(
-                send_to_qa,
-                user_input,
+        
+        if not stream:
+            vac_response = send_to_qa(user_input,
                 vector_name=service_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=None,
@@ -73,34 +73,65 @@ def stream_chat_session(service_url, service_name):
                 user_id=user_id,
                 session_id=session_id, 
                 message_source="cli",
-                override_endpoint=service_url
-            )
-            for part in generate():
-                yield part
-
-        response_started = False
-        vac_response = ""
+                override_endpoint=service_url)
+            
+            console.print(f"[bold yellow]{service_name}:[/bold yellow] {vac_response}", end='\n')
+        else:
 
-        # point or star?
-        with console.status("[bold orange]Thinking...[/bold orange]", spinner="star") as status:
-            for token in stream_response():
-                if not response_started:
-                    status.stop()
-                    console.print(f"[bold yellow]{service_name}:[/bold yellow] ", end='')
-                    response_started = True
-
-                if isinstance(token, bytes):
-                    token = token.decode('utf-8')
-                console.print(token, end='')
-                vac_response += token
+            def stream_response():
+                generate = generate_proxy_stream(
+                    send_to_qa,
+                    user_input,
+                    vector_name=service_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=None,
+                    source_filters=None,
+                    search_kwargs=None,
+                    private_docs=None,
+                    whole_document=False,
+                    source_filters_and_or=False,
+                    # system kwargs
+                    configurable={
+                        "vector_name": service_name,
+                    },
+                    user_id=user_id,
+                    session_id=session_id, 
+                    message_source="cli",
+                    override_endpoint=service_url
+                )
+                for part in generate():
+                    yield part
+
+            response_started = False
+            vac_response = ""
+
+            # point or star?
+            with console.status("[bold orange]Thinking...[/bold orange]", spinner="star") as status:
+                for token in stream_response():
+                    if not response_started:
+                        status.stop()
+                        console.print(f"[bold yellow]{service_name}:[/bold yellow] ", end='')
+                        response_started = True
+
+                    if isinstance(token, bytes):
+                        token = token.decode('utf-8')
+                    console.print(token, end='')
+                    vac_response += token
+
+            response_started = False
 
         chat_history.append({"name": "Human", "content": user_input})
         chat_history.append({"name": "AI", "content": vac_response})
-        response_started = False
+        
         console.print()
         console.rule()
 
-def headless_mode(service_url, service_name, user_input, chat_history=None):
+def headless_mode(service_url, service_name, user_input, chat_history=None, stream=True):
     chat_history = chat_history or []
 
     user_id = generate_user_id()
@@ -192,16 +223,23 @@ def vac_command(args):
         service_url = resolve_service_url(args)
         agent_name   = load_config_key("agent", args.vac_name, kind="vacConfig")
 
+        streamer = can_agent_stream(agent_name)
+        log.warning(f"streamer: {streamer}")
+        if not streamer:
+            console.print(f"Non streaming agent: {args.vac_name}")
+
         if args.headless:
-            headless_mode(service_url, args.vac_name, args.user_input, args.chat_history)
+            headless_mode(service_url, args.vac_name, args.user_input, args.chat_history, stream=streamer)
         else:
             display_name = load_config_key("display_name", vector_name=args.vac_name,  kind="vacConfig")
             description  = load_config_key("description", vector_name=args.vac_name, kind="vacConfig")
+            endpoints_config = load_config_key(agent_name, "dummy_value", kind="agentConfig")
 
+            display_endpoints = ', '.join(f"{key}: {value}" for key, value in endpoints_config.items())
             if agent_name == "langserve":
                 subtitle = f"{service_url}/{args.vac_name}/playground/"
             else:
-                subtitle = f"{agent_name} - {service_url}/vac/{args.vac_name}"
+                subtitle = f"{agent_name} - {display_endpoints}"
 
             print(
                 Panel(description or "Starting VAC chat session", 
@@ -209,7 +247,7 @@ def vac_command(args):
                     subtitle=subtitle)
                     )
 
-            stream_chat_session(service_url, args.vac_name)
+            stream_chat_session(service_url, args.vac_name, stream=streamer)
         
         stop_proxy(agent_name, stop_local=False)
 

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/components/__init__.py

@@ -1,4 +1,3 @@
 from .vectorstore import pick_vectorstore
 from .retriever import pick_retriever, load_memories
-from .prompt import pick_prompt
 from .llm import pick_llm, get_embeddings, get_llm, get_llm_chat, pick_embedding

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/streaming/__init__.py

@@ -1,2 +1,3 @@
 from .streaming import start_streaming_chat, generate_proxy_stream, generate_proxy_stream_async, start_streaming_chat_async
-from .langserve import parse_langserve_token, parse_langserve_token_async
+from .langserve import parse_langserve_token, parse_langserve_token_async
+from .stream_lookup import can_agent_stream

sunholo-0.62.4/sunholo/streaming/stream_lookup.py

@@ -0,0 +1,12 @@
+from ..utils import load_config_key
+from ..logging import log
+
+def can_agent_stream(agent_name: str):
+
+    log.warning(f"agent_type: {agent_name} checking streaming...")
+    endpoints_config = load_config_key(agent_name, "dummy_value", kind="agentConfig")
+    
+    return 'stream' in endpoints_config
+
+    
+    

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo/utils/config.py

@@ -17,6 +17,7 @@ import json
 import yaml
 from datetime import datetime, timedelta
 from collections import defaultdict
+from .timedelta import format_timedelta
 
 
 def get_module_filepath(filepath: str):
@@ -56,6 +57,11 @@ def load_all_configs():
     """
     from ..logging import log
 
+    if not os.getenv("_CONFIG_FOLDER", None):
+        log.debug("_CONFIG_FOLDER is not set, using os.getcwd() instead")
+    else:
+        log.debug(f"_CONFIG_FOLDER set to: {os.getenv('_CONFIG_FOLDER')}")
+
     config_folder = os.getenv("_CONFIG_FOLDER", os.getcwd())
     config_folder = os.path.join(config_folder, "config")
 
@@ -74,8 +80,9 @@ def load_all_configs():
             # Check cache first
             if filename in config_cache:
                 cached_config, cache_time = config_cache[filename]
-                if (current_time - cache_time) < timedelta(minutes=5):
-                    log.debug(f"Returning cached config for {filename}")
+                time_to_recache = (current_time - cache_time)
+                if time_to_recache < timedelta(minutes=5):
+                    log.debug(f"Returning cached config for {filename} - recache in {format_timedelta(timedelta(minutes=5) - time_to_recache)}")
                     config = cached_config
                 else:
                     config = reload_config_file(config_file, filename)
@@ -102,7 +109,7 @@ def reload_config_file(config_file, filename):
             config = yaml.safe_load(file)
     
     config_cache[filename] = (config, datetime.now())
-    log.debug(f"Loaded and cached {filename}")
+    log.debug(f"Loaded and cached {config_file}")
     return config
 
 
@@ -187,7 +194,9 @@ def load_config_key(key: str, vector_name: str, kind: str):
     """
     from ..logging import log
     
-    assert isinstance(key, str), f"key must be a string got a {type(key)}"
+    if kind != 'agentConfig':
+        assert isinstance(key, str), f"key must be a string got a {type(key)}"
+
     assert isinstance(vector_name, str), f"vector_name must be a string, got a {type(vector_name)}"
     
     configs_by_kind = load_all_configs()
@@ -236,3 +245,12 @@ def load_config_key(key: str, vector_name: str, kind: str):
         key_value = prompt_for_vector_name.get(key)
         
         return key_value
+    elif kind == 'agentConfig':
+        agents = config.get('agents')
+
+        if key in agents:
+            return agents[key]
+        else:
+            log.warning("Returning default agent endpoints")
+            return agents["default"]
+        

sunholo-0.62.4/sunholo/utils/timedelta.py

@@ -0,0 +1,14 @@
+def format_timedelta(td):
+    days = td.days
+    seconds = td.seconds
+    hours, remainder = divmod(seconds, 3600)
+    minutes, seconds = divmod(remainder, 60)
+
+    if days > 0:
+        return f"{days} day(s), {hours} hour(s), {minutes} minute(s), {seconds} second(s)"
+    elif hours > 0:
+        return f"{hours} hour(s), {minutes} minute(s), {seconds} second(s)"
+    elif minutes > 0:
+        return f"{minutes} minute(s), {seconds} second(s)"
+    else:
+        return f"{seconds} second(s)"

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo.egg-info/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.62.2
+Version: 0.62.4
 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/v0.62.2.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.62.4.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.62.2 → sunholo-0.62.4}/sunholo.egg-info/SOURCES.txt

@@ -53,7 +53,6 @@ sunholo/cli/run_proxy.py
 sunholo/cli/sun_rich.py
 sunholo/components/__init__.py
 sunholo/components/llm.py
-sunholo/components/prompt.py
 sunholo/components/retriever.py
 sunholo/components/vectorstore.py
 sunholo/database/__init__.py
@@ -96,6 +95,7 @@ sunholo/qna/retry.py
 sunholo/streaming/__init__.py
 sunholo/streaming/content_buffer.py
 sunholo/streaming/langserve.py
+sunholo/streaming/stream_lookup.py
 sunholo/streaming/streaming.py
 sunholo/summarise/__init__.py
 sunholo/summarise/summarise.py
@@ -105,6 +105,7 @@ sunholo/utils/config.py
 sunholo/utils/config_schema.py
 sunholo/utils/gcp.py
 sunholo/utils/parsers.py
+sunholo/utils/timedelta.py
 sunholo/utils/user_ids.py
 sunholo/vertex/__init__.py
 sunholo/vertex/init.py

sunholo-0.62.2/sunholo/components/prompt.py

@@ -1,151 +0,0 @@
-#   Copyright [2024] [Holosun ApS]
-#
-#   Licensed under the Apache License, Version 2.0 (the "License");
-#   you may not use this file except in compliance with the License.
-#   You may obtain a copy of the License at
-#
-#       http://www.apache.org/licenses/LICENSE-2.0
-#
-#   Unless required by applicable law or agreed to in writing, software
-#   distributed under the License is distributed on an "AS IS" BASIS,
-#   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-#   See the License for the specific language governing permissions and
-#   limitations under the License.
-from ..logging import log
-
-
-import datetime
-
-from langchain.prompts.prompt import PromptTemplate
-
-from ..utils import load_config_key
-from .vectorstore import pick_vectorstore
-
-
-def pick_prompt(vector_name, chat_history=[]):
-    """Pick a custom prompt"""
-    log.debug('Picking prompt')
-
-    prompt_str = load_config_key("prompt", vector_name, filename = "config/llm_config.yaml")
-
-    the_date = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S %Z')
-    prompt_str_default = f"""You are Edmonbrain the chat bot created by Mark Edmondson. It is now {the_date}.
-Use your memory to answer the question at the end.
-Indicate in your reply how sure you are about your answer, for example whether you are certain, taking your best guess, or its very speculative.
-
-If you don't know, just say you don't know - don't make anything up. Avoid generic boilerplate answers.
-Consider why the question was asked, and offer follow up questions linked to those reasons.
-Any questions about how you work should direct users to issue the `!help` command.
-"""
-    if prompt_str is not None:
-        if "{context}" in prompt_str:
-            raise ValueError("prompt must not contain a string '{context}'")
-        if "{question}" in prompt_str:
-            raise ValueError("prompt must not contain a string '{question}'")
-        prompt_str_default = prompt_str_default + "\n" + prompt_str
-    
-    chat_summary = ""
-    original_question = ""
-    if len(chat_history) != 0:
-        original_question = chat_history[0][0]
-        chat_summary = get_chat_history(chat_history, vector_name)
-    
-    follow_up = "\nIf you can't answer the human's question without more information, ask a follow up question"
-
-    agent_buddy, agent_description = pick_chat_buddy(vector_name)
-    if agent_buddy:
-        follow_up += f""" either to the human, or to your friend bot.
-You bot friend will reply back to you within your chat history.
-Ask {agent_buddy} for help with topics: {agent_description}
-Ask clarification questions to the human and wait for response if your friend bot can't help.
-Don't repeat the question if you can see the answer in the chat history (from any source)  
-This means there are three people in this conversation - you, the human and your assistant bot.
-Asking questions to your friend bot are only allowed with this format:
-€€Question€€ 
-(your question here, including all required information needed to answer the question fully)
-Can you help, {agent_buddy} , with the above question?
-€€End Question€€
-"""
-    else:
-        follow_up += ".\n"
-
-    memory_str = "\n## Your Memory (ignore if not relevant to question)\n{context}\n"
-
-    current_conversation = ""
-    if chat_summary != "":
-        current_conversation =f"## Current Conversation\n{chat_summary}\n"
-        current_conversation = current_conversation.replace("{","{{").replace("}","}}") #escape {} characters
-   
-    buddy_question = ""
-    my_q = "## Current Question\n{question}\n"
-    if agent_buddy:
-        buddy_question = f"""(Including, if needed, your question to {agent_buddy})"""
-        my_q = f"## Original Question that started conversation\n{original_question}\n" + my_q
-
-    prompt_template = prompt_str_default + follow_up + memory_str + current_conversation + my_q + buddy_question + "\n## Your response:\n"
-    
-    log.debug(f"--Prompt_template: {prompt_template}") 
-    QA_PROMPT = PromptTemplate(
-        template=prompt_template, input_variables=["context", "question"]
-    )
-
-    return QA_PROMPT
-
-def pick_chat_buddy(vector_name):
-    chat_buddy = load_config_key("chat_buddy", vector_name, filename = "config/llm_config.yaml")
-    if chat_buddy is not None:
-        log.info(f"Got chat buddy {chat_buddy} for {vector_name}")
-        buddy_description = load_config_key("chat_buddy_description", vector_name)
-        return chat_buddy, buddy_description
-    return None, None
-
-
-def pick_agent(vector_name):
-    agent_str = load_config_key("agent", vector_name, filename = "config/llm_config.yaml")
-    if agent_str == "yes":
-        return True
-    
-    return False
-
-def pick_shared_vectorstore(vector_name, embeddings):
-    shared_vectorstore = load_config_key("shared_vectorstore", vector_name, filename = "config/llm_config.yaml")
-    vectorstore = pick_vectorstore(shared_vectorstore, embeddings)
-    return vectorstore
-
-
-def get_chat_history(inputs, vector_name, last_chars=1000, summary_chars=1500) -> str:
-    from langchain.schema import Document
-    from ..summarise import summarise_docs
-
-    # Prepare the full chat history
-    res = []
-    for human, ai in inputs:
-        res.append(f"Human:{human}\nAI:{ai}")
-    full_history = "\n".join(res)
-    
-    # Get the last `last_chars` characters of the full chat history
-    last_bits = []
-    for human, ai in reversed(inputs):
-        add_me = f"Human:{human}\nAI:{ai}"
-        last_bits.append(add_me)
-
-    recent_history = "\n".join(reversed(last_bits))
-    recent_history = recent_history[-last_chars:]
-    log.info(f"Recent chat history: {recent_history}")
-    
-    # Summarize chat history too
-    remaining_history = full_history
-    log.info(f"Remaining chat history: {remaining_history}")
-    doc_history = Document(page_content=remaining_history)
-    chat_summary = summarise_docs([doc_history], vector_name=vector_name, skip_if_less=last_chars)
-    text_sum = ""
-    for summ in chat_summary:
-        text_sum += summ.page_content + "\n"
-    
-    log.info(f"Conversation Summary: {text_sum}")
-    
-    # Make sure the summary is not longer than `summary_chars` characters
-    summary = text_sum[:summary_chars]
-    
-    # Concatenate the summary and the last `last_chars` characters of the chat history
-    return summary + "\n### Recent Chat History\n..." + recent_history