Update config.py

config.py

@@ -1,5 +1,3 @@
-"""Configuration for Veda Programming Assistant with Distillation"""
-
 import os
 
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
@@ -11,7 +9,7 @@ os.makedirs(MODEL_DIR, exist_ok=True)
 
 DATABASE_PATH = os.path.join(DATA_DIR, "conversations.db")
 
-# Model settings
+# Student model settings
 VOCAB_SIZE = 8000
 MAX_LENGTH = 512
 D_MODEL = 256
@@ -20,28 +18,17 @@ NUM_LAYERS = 4
 FF_DIM = 512
 BATCH_SIZE = 4
 
-# Generation defaults
-DEFAULT_TEMPERATURE = 0.7
-DEFAULT_MAX_TOKENS = 200
-
-# ====== DISTILLATION SETTINGS ======
-# OpenRouter API
-OPENROUTER_API_KEY = os.environ.get(
-    "OPENROUTER_API_KEY",
-    "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"
-)
+# OpenRouter Teacher
+OPENROUTER_API_KEY = os.environ.get("OPENROUTER_API_KEY", "").strip()
 OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1/chat/completions"
 
-# Teacher model (free tier)
-TEACHER_MODEL = "cognitivecomputations/dolphin-mistral-24b-venice-edition:free"
+# ✅ Teacher model (confirmed working)
+TEACHER_MODEL = "mistralai/mistral-7b-instruct:free"
 
-# Distillation settings
-DISTILLATION_ENABLED = True
-AUTO_LEARN_FROM_TEACHER = True  # Automatically save teacher responses for training
-MIN_SAMPLES_FOR_DISTILL_TRAINING = 20  # Minimum teacher samples before retraining
-TEACHER_TEMPERATURE = 0.7
-TEACHER_MAX_TOKENS = 500
+# Optional fallback if that model ever fails
+TEACHER_FALLBACK_MODELS = [
+    "openrouter/auto"
+]
 
-# When to ask teacher (confidence threshold)
-# If student response is too short or seems low quality, ask teacher
-MIN_RESPONSE_LENGTH = 20  # If response shorter than this, ask teacher
+TEACHER_TEMPERATURE = 0.7
+TEACHER_MAX_TOKENS = 400