diff options
| -rw-r--r-- | bot.py | 13 | ||||
| -rw-r--r-- | main.py | 13 |
2 files changed, 10 insertions, 16 deletions
@@ -18,7 +18,8 @@ def ignore_connection_closed_ok(loop, context): #this doesn't fucking work loop.default_exception_handler(context) -system_prompt = "you are a funny discord user that uses discrd" +def load_system_prompt() -> str: + return Path("system_prompt.txt").read_text(encoding="utf-8") FILES_DIR = Path("files") LOG_DIR = FILES_DIR / "log" @@ -91,13 +92,13 @@ def load_longterm() -> list[str]: log_error(f"[{timestamp()}] error reading longterm: {e}") return [] -def build_payload(sys_prompt: str, user_input: str, rebuild: bool = False) -> str: +def build_payload(user_input: str, rebuild: bool = False) -> str: global vector_db_ready should_rebuild = rebuild or not vector_db_ready # always rebuild on first ever message if should_rebuild: vector_db_ready = True return json.dumps({ - "system_prompt": sys_prompt, + "system_prompt": load_system_prompt(), "user_input": user_input, "shortterm": load_shortterm(), "longterm": load_longterm(), @@ -184,7 +185,6 @@ async def random_dm(): continue victim = random.choice(members) payload = build_payload( - system_prompt, f"System: you are sending a dm to {victim}, write a long, playful story or an over-the-top pun monologue about anything. " "It should sound like something faggot-chan would post when she's bored — chaotic, funny, and full of personality (at least 20 words)." "keep in mind this is a dm so make it a bit personal" @@ -215,7 +215,6 @@ async def random_message(): channel = discord.utils.get(guild.text_channels, name="general") if channel: payload = build_payload( - system_prompt, "System: write a long, playful story or an over-the-top pun monologue about cheese. or anything random" "it should sound like something faggot-chan would post when she's bored." "chaotic, funny, and full of personality (at least 15 words and no more than 20)" @@ -361,7 +360,7 @@ async def on_message(message): user_input = f'{message.author} replied to {replied_message.author} saying: "{replied_content}" with reply: "{content}"' else: user_input = f"{message.author} said in DM: {content}" - payload = build_payload(system_prompt, user_input, rebuild) + payload = build_payload(user_input, rebuild) stop_typing = asyncio.Event() asyncio.create_task(keep_typing(message.channel, stop_typing)) ok, result = await call_main(payload) @@ -392,7 +391,7 @@ async def on_message(message): else: user_input = f"{message.author} said: {content}" - payload = build_payload(system_prompt, user_input, rebuild) + payload = build_payload(user_input, rebuild) stop_typing = asyncio.Event() asyncio.create_task(keep_typing(message.channel, stop_typing)) ok, result = await call_main(payload) @@ -4,13 +4,13 @@ import json import re # thank you claude import ollama import websockets +from pathlib import Path from websockets.server import serve from datetime import datetime chatmodel = "hf.co/Stetsoncody/Llama-3.2-3B-Instruct-abliterated-GGUF:Q4_K_M" embedmodel = 'hf.co/CompendiumLabs/bge-base-en-v1.5-gguf' -NUM_THREADS = 6 # hardcoded since threads.txt is gone, change if needed def times() -> str: return datetime.now().strftime("%Y-%m-%d %H:%M:%S") @@ -151,17 +151,12 @@ async def ollamachat(websocket): messages.insert(0, {"role": "system", "content": system_prompt + rag_block}) + params = json.loads(Path("config/params.json").read_text()) + response = ollama.chat( model=chatmodel, messages=messages, - options={ - "temperature": 0.83, - "top_p": 0.9, - "top_k": 40, - "repeat_penalty": 1.1, - "num_predict": 1024, - "num_thread": NUM_THREADS, - } + options=params, ) # strip <think> blocks in case the model ignores think=False, thank you claude |