made it so it reads the system prompt and parameters from a file (this was vibecoded but it works so it's okay)

1 files changed, 4 insertions, 9 deletions

diff --git a/main.py b/main.py
index 74533f5..ad5dc4f 100644
--- a/main.py
+++ b/main.py
@@ -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