Auto-update: Thu Aug 8 17:24:51 PDT 2024

sijapi/__init__.py

@@ -78,10 +78,6 @@ DAY_SHORT_FMT = os.getenv("DAY_SHORT_FMT")
 # Large language model
 LLM_URL = os.getenv("LLM_URL", "http://localhost:11434")
 LLM_SYS_MSG = os.getenv("SYSTEM_MSG", "You are a helpful AI assistant.")
-DEFAULT_LLM = os.getenv("DEFAULT_LLM", "llama3")
-DEFAULT_VISION = os.getenv("DEFAULT_VISION", "llava")
-DEFAULT_VOICE = os.getenv("DEFAULT_VOICE", "Luna")
-DEFAULT_11L_VOICE = os.getenv("DEFAULT_11L_VOICE", "Victoria")
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
 
 # Summarization
@@ -118,8 +114,6 @@ TTS_DIR = DATA_DIR / "tts"
 os.makedirs(TTS_DIR, exist_ok=True)
 VOICE_DIR = TTS_DIR / 'voices'
 os.makedirs(VOICE_DIR, exist_ok=True)
-PODCAST_DIR = os.getenv("PODCAST_DIR", TTS_DIR / "sideloads")
-os.makedirs(PODCAST_DIR, exist_ok=True)
 TTS_OUTPUT_DIR = TTS_DIR / 'outputs'
 os.makedirs(TTS_OUTPUT_DIR, exist_ok=True)
 TTS_SEGMENTS_DIR = TTS_DIR / 'segments'

sijapi/config/.env-example

@@ -323,8 +323,6 @@ MS365_REDIRECT_PATH=¿SECRET?      # <--- e.g. http://localhost:4444/MS365/oauth
 #─── via comfyui (stable diffusion): ─────── S̝͖̦͓̪̻ O̡͖̘̫͇̟   H̢͔͔̫͉͜ O̢̢͉̞͍̘ T̟͍͍̪̦̞  R I G H T   N O W 
 LLM_URL=http://localhost:11434
 SYSTEM_MSG=You are a helpful AI assistant.
-DEFAULT_LLM=dolphin-mistral
-DEFAULT_VISION=llava-llama3
 OPENAI_API_KEY=¿SECRET?                # <--- not presently implemented for anything
 SUMMARY_MODEL='command-r:latest'
 SUMMARY_CHUNK_SIZE=16384
@@ -335,7 +333,6 @@ SUMMARY_MIN_LENGTH=64
 SUMMARY_TOKEN_LIMIT=16384
 SUMMARY_INSTRUCT='You are an AI assistant that provides accurate summaries of text -- nothing more and nothing less. You must not include ANY extraneous text other than the sumary. Do not include comments apart from the summary, do not preface the summary, and do not provide any form of postscript. Do not add paragraph breaks. Do not add any kind of formatting. Your response should begin with, consist of, and end with an accurate plaintext summary.'
 SUMMARY_INSTRUCT_TTS='You are an AI assistant that summarizes emails -- nothing more and nothing less. You must not include ANY extraneous text other than the sumary. Do not include comments apart from the summary, do not preface the summary, and do not provide any form of postscript. Do not add paragraph breaks. Do not add any kind of formatting. Your response should begin with, consist of, and end with an accurate plaintext summary. Your response will undergo Text-To-Speech conversion and added to Sanjays private podcast. Providing adequate context (Sanjay did not send this question to you, he will only hear your response) but aiming for conciseness and precision, and bearing in mind the Text-To-Speech conversion (avoiding acronyms and formalities), summarize the following.'
-DEFAULT_VOICE=joanne
 WHISPER_CPP_DIR='whisper.cpp'
 WHISPER_CPP_MODELS=tiny,base,base-en,small,medium,medium-en,large-v3
 WEBCLIPPER_TTS=elevenlabs
@@ -351,9 +348,6 @@ DAY_SHORT_FMT="%Y-%m-%d"
 #   designed for use with `ollama`, but most of the functionality should be equal with
 #   LM Studio, LocalAI, ect...                                                            
 #   
-#   DEFAULT_LLM is self-explanatory; DEFAULT_VISION is used for image recognition within
-#   a multimodal chat context, such as on the ig module for generating intelligible      
-#   comments to Instagram posts, or more realistic captions for img-generated images.     
 #   
 #   Note it's possible to specify a separate model for general purposes and for 
 #   summarization tasks. The other SUMMARY_ variables call for some explanation, 
@@ -391,8 +385,6 @@ DAY_SHORT_FMT="%Y-%m-%d"
 #                           tts output was requested; tends to yield "cleaner" audio 
 #                           with less numbers (page numbers, citations) and other 
 #                           information extraneous to spoken contexts.
-#                           
-#           DEFAULT_VOICE:  used for all tts tasks when a specific voice is not requested.
 #                                                                                         
 #                                      ──────────                                         
 #                                                                                         

sijapi/routers/archivist.py

@@ -30,7 +30,7 @@ from urllib3.util.retry import Retry
 from datetime import datetime as dt_datetime
 from better_profanity import profanity
 from sijapi.utilities import html_to_markdown, sanitize_filename, assemble_journal_path, assemble_archive_path, contains_profanity, is_ad_or_tracker, initialize_adblock_rules, contains_blacklisted_word
-from sijapi import L, API, Archivist, BLOCKLISTS_DIR, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR, DEFAULT_11L_VOICE, DEFAULT_VOICE
+from sijapi import L, API, Archivist, BLOCKLISTS_DIR, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR
 
 archivist = APIRouter()
 

sijapi/routers/email.py

@@ -22,7 +22,7 @@ import ssl
 import yaml
 from typing import List, Dict, Optional, Set
 from datetime import datetime as dt_datetime
-from sijapi import L, PODCAST_DIR, DEFAULT_VOICE, EMAIL_CONFIG, EMAIL_LOGS
+from sijapi import L, Dir, EMAIL_CONFIG, EMAIL_LOGS
 from sijapi.routers import gis, img, tts, llm
 from sijapi.utilities import clean_text, assemble_journal_path, extract_text, prefix_lines
 from sijapi.classes import EmailAccount, IMAPConfig, SMTPConfig, IncomingEmail, EmailContact, AutoResponder
@@ -205,7 +205,7 @@ async def summarize_single_email(this_email: IncomingEmail, podcast: bool = Fals
         attachment_texts = await extract_attachments(this_email.attachments)
         email_content += "\n—--\n" + "\n—--\n".join([f"Attachment: {text}" for text in attachment_texts])
     summary = await llm.summarize_text(email_content)
-    await tts.local_tts(text_content = summary, speed = 1.1, voice = DEFAULT_VOICE, podcast = podcast, output_path = tts_path)
+    await tts.local_tts(text_content = summary, speed = 1.1, voice = Tts.xtts.default, podcast = podcast, output_path = tts_path)
     md_summary = f'```ad.summary\n'
     md_summary += f'title: {this_email.subject}\n'
     md_summary += f'{summary}\n'

sijapi/routers/llm.py

@@ -26,7 +26,7 @@ import tempfile
 import shutil
 import html2text
 import markdown
-from sijapi import L, LLM_SYS_MSG, DEFAULT_LLM, DEFAULT_VISION, REQUESTS_DIR, OBSIDIAN_CHROMADB_COLLECTION, OBSIDIAN_VAULT_DIR, DOC_DIR, OPENAI_API_KEY, DEFAULT_VOICE, SUMMARY_INSTRUCT, SUMMARY_CHUNK_SIZE, SUMMARY_TPW, SUMMARY_CHUNK_OVERLAP, SUMMARY_LENGTH_RATIO, SUMMARY_TOKEN_LIMIT, SUMMARY_MIN_LENGTH, SUMMARY_MODEL
+from sijapi import L, LLM_SYS_MSG, REQUESTS_DIR, OBSIDIAN_CHROMADB_COLLECTION, OBSIDIAN_VAULT_DIR, DOC_DIR, OPENAI_API_KEY, SUMMARY_INSTRUCT, SUMMARY_CHUNK_SIZE, SUMMARY_TPW, SUMMARY_CHUNK_OVERLAP, SUMMARY_LENGTH_RATIO, SUMMARY_TOKEN_LIMIT, SUMMARY_MIN_LENGTH, SUMMARY_MODEL
 from sijapi.utilities import convert_to_unix_time, sanitize_filename, ocr_pdf, clean_text, should_use_ocr, extract_text_from_pdf, extract_text_from_docx, read_text_file, str_to_bool, get_extension
 from sijapi.routers import tts
 from sijapi.routers.asr import transcribe_audio
@@ -87,7 +87,7 @@ async def generate_response(prompt: str):
     return {"response": output['response']}
 
 
-async def query_ollama(usr: str, sys: str = LLM_SYS_MSG, model: str = DEFAULT_LLM, max_tokens: int = 200):
+async def query_ollama(usr: str, sys: str = LLM_SYS_MSG, model: str = Llm.chat.model, max_tokens: int = 200):
     messages = [{"role": "system", "content": sys},
                 {"role": "user", "content": usr}]
     LLM = Ollama()
@@ -105,7 +105,7 @@ async def query_ollama(usr: str, sys: str = LLM_SYS_MSG, model: str = DEFAULT_LL
 async def query_ollama_multishot(
     message_list: List[str],
     sys: str = LLM_SYS_MSG,
-    model: str = DEFAULT_LLM,
+    model: str = Llm.chat.model,
     max_tokens: int = 200
 ):
     if len(message_list) % 2 == 0:
@@ -231,9 +231,9 @@ async def stream_messages_with_vision(message: dict, model: str, num_predict: in
         
 def get_appropriate_model(requested_model):
     if requested_model == "gpt-4-vision-preview":
-        return DEFAULT_VISION
+        return Llm.vision.model
     elif not is_model_available(requested_model):
-        return DEFAULT_LLM
+        return Llm.chat.model
     else:
         return requested_model
 
@@ -314,7 +314,7 @@ async def chat_completions_options(request: Request):
             ],
             "created": int(time.time()),
             "id": str(uuid.uuid4()),
-            "model": DEFAULT_LLM,
+            "model": Llm.chat.model,
             "object": "chat.completion.chunk",
         },
         status_code=200,
@@ -533,7 +533,7 @@ async def summarize_tts_endpoint(
     instruction: str = Form(SUMMARY_INSTRUCT),
     file: Optional[UploadFile] = File(None),
     text: Optional[str] = Form(None),
-    voice: Optional[str] = Form(DEFAULT_VOICE),
+    voice: Optional[str] = Form(None),
     speed: Optional[float] = Form(1.2),
     podcast: Union[bool, str] = Form(False)
 ):
@@ -577,7 +577,7 @@ async def summarize_tts_endpoint(
 async def summarize_tts(
     text: str, 
     instruction: str = SUMMARY_INSTRUCT,
-    voice: Optional[str] = DEFAULT_VOICE,
+    voice: Optional[str] = None,
     speed: float = 1.1,
     podcast: bool = False,
     LLM: Ollama = None

sijapi/routers/news.py

@@ -18,7 +18,7 @@ from markdownify import markdownify as md
 from better_profanity import profanity
 from fastapi import APIRouter, BackgroundTasks, UploadFile, Form, HTTPException, Query, Path as FastAPIPath
 from pathlib import Path
-from sijapi import L, News, Archivist, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR, DEFAULT_11L_VOICE, DEFAULT_VOICE
+from sijapi import L, News, Archivist, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR
 from sijapi.utilities import html_to_markdown, download_file, sanitize_filename, assemble_journal_path, assemble_archive_path, contains_profanity, is_ad_or_tracker
 from sijapi.routers import gis, llm, tts, note
 
@@ -37,7 +37,7 @@ async def clip_post(
     url: str = Form(...),
     title: Optional[str] = Form(None),
     tts: str = Form('summary'),
-    voice: str = Form(DEFAULT_VOICE),
+    voice: str = Form(None),
 ):
     result = await process_and_save_article(bg_tasks, url, title, tts, voice)
     return {"message": "Clip saved successfully", "result": result}
@@ -46,7 +46,7 @@ async def clip_post(
 async def clip_get(
     bg_tasks: BackgroundTasks,
     url: str,
-    voice: str = Query(DEFAULT_VOICE)
+    voice: str = Query(None)
 ):
     result = await process_and_save_article(bg_tasks, url, None, tts, voice)
     return {"message": "Clip saved successfully", "result": result}
@@ -124,7 +124,7 @@ async def process_news_site(site, bg_tasks: BackgroundTasks):
                 earliest_date, 
                 bg_tasks, 
                 tts_mode=site.tts if hasattr(site, 'tts') else "off",
-                voice=site.voice if hasattr(site, 'voice') else DEFAULT_11L_VOICE
+                voice=site.voice if hasattr(site, 'voice') else Tts.elevenlabs.default
             ))
             tasks.append(task)
         
@@ -136,7 +136,7 @@ async def process_news_site(site, bg_tasks: BackgroundTasks):
         err(f"Error processing {site.name}: {str(e)}")
 
 
-async def download_and_save_article(article, site_name, earliest_date, bg_tasks: BackgroundTasks, tts_mode: str = "off", voice: str = DEFAULT_11L_VOICE):
+async def download_and_save_article(article, site_name, earliest_date, bg_tasks: BackgroundTasks, tts_mode: str = "off", voice: str = Tts.elevenlabs.default):
     try:
         url = article.url
         parsed_article = await fetch_and_parse_article(url)
@@ -156,7 +156,7 @@ async def process_and_save_article(
     url: str,
     title: Optional[str] = None,
     tts_mode: str = "summary",
-    voice: str = DEFAULT_VOICE,
+    voice: str = Tts.elevenlabs.default,
     site_name: Optional[str] = None
 ) -> str:
     

sijapi/routers/note.py

@@ -17,7 +17,7 @@ from dateutil.parser import parse as dateutil_parse
 from fastapi import HTTPException, status
 from pathlib import Path
 from fastapi import APIRouter, Query, HTTPException
-from sijapi import API, L, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR, OBSIDIAN_BANNER_SCENE, DEFAULT_11L_VOICE, DEFAULT_VOICE, GEO
+from sijapi import API, L, OBSIDIAN_VAULT_DIR, OBSIDIAN_RESOURCES_DIR, OBSIDIAN_BANNER_SCENE, GEO
 from sijapi.routers import asr, cal, gis, img, llm, serve, timing, tts, weather
 from sijapi.utilities import assemble_journal_path, convert_to_12_hour_format, sanitize_filename, convert_degrees_to_cardinal, check_file_name, HOURLY_COLUMNS_MAPPING
 from sijapi.classes import Location
@@ -122,7 +122,7 @@ async def process_document(
     document: File,
     title: Optional[str] = None,
     tts_mode: str = "summary",
-    voice: str = DEFAULT_VOICE
+    voice: str = None,
 ):
     timestamp = dt_datetime.now().strftime('%b %d, %Y at %H:%M')
 

sijapi/routers/tts.py

@@ -488,10 +488,10 @@ def copy_to_podcast_dir(file_path):
         # Extract the file name from the file path
         file_name = Path(file_path).name
         
-        # Construct the destination path in the PODCAST_DIR
+        # Construct the destination path in the podcast folder
         destination_path = Path(Dir.PODCAST) / file_name
         
-        # Copy the file to the PODCAST_DIR
+        # Copy the file to the podcast folder
         shutil.copy(file_path, destination_path)
         
         print(f"File copied successfully to {destination_path}")