vedaco commited on
Commit
721b9cf
·
verified ·
1 Parent(s): c8e91c6

Update config.py

Browse files
Files changed (1) hide show
  1. config.py +26 -2
config.py CHANGED
@@ -1,4 +1,4 @@
1
- """Configuration for Veda Programming Assistant"""
2
 
3
  import os
4
 
@@ -11,6 +11,7 @@ os.makedirs(MODEL_DIR, exist_ok=True)
11
 
12
  DATABASE_PATH = os.path.join(DATA_DIR, "conversations.db")
13
 
 
14
  VOCAB_SIZE = 8000
15
  MAX_LENGTH = 512
16
  D_MODEL = 256
@@ -19,5 +20,28 @@ NUM_LAYERS = 4
19
  FF_DIM = 512
20
  BATCH_SIZE = 4
21
 
 
22
  DEFAULT_TEMPERATURE = 0.7
23
- DEFAULT_MAX_TOKENS = 200
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Configuration for Veda Programming Assistant with Distillation"""
2
 
3
  import os
4
 
 
11
 
12
  DATABASE_PATH = os.path.join(DATA_DIR, "conversations.db")
13
 
14
+ # Model settings
15
  VOCAB_SIZE = 8000
16
  MAX_LENGTH = 512
17
  D_MODEL = 256
 
20
  FF_DIM = 512
21
  BATCH_SIZE = 4
22
 
23
+ # Generation defaults
24
  DEFAULT_TEMPERATURE = 0.7
25
+ DEFAULT_MAX_TOKENS = 200
26
+
27
+ # ====== DISTILLATION SETTINGS ======
28
+ # OpenRouter API
29
+ OPENROUTER_API_KEY = os.environ.get(
30
+ "OPENROUTER_API_KEY",
31
+ "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"
32
+ )
33
+ OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1/chat/completions"
34
+
35
+ # Teacher model (free tier)
36
+ TEACHER_MODEL = "cognitivecomputations/dolphin-mistral-24b-venice-edition:free"
37
+
38
+ # Distillation settings
39
+ DISTILLATION_ENABLED = True
40
+ AUTO_LEARN_FROM_TEACHER = True # Automatically save teacher responses for training
41
+ MIN_SAMPLES_FOR_DISTILL_TRAINING = 20 # Minimum teacher samples before retraining
42
+ TEACHER_TEMPERATURE = 0.7
43
+ TEACHER_MAX_TOKENS = 500
44
+
45
+ # When to ask teacher (confidence threshold)
46
+ # If student response is too short or seems low quality, ask teacher
47
+ MIN_RESPONSE_LENGTH = 20 # If response shorter than this, ask teacher