Change default model from MODEL_1 to MODEL_0 to match stable config

Respects existing stable configuration where:
- MODEL_0 = Hermes (default for classification and feedback)
- MODEL_1 = Qwen (for code generation)
- MODEL_2 = GPT

Updated:
- All function defaults in activity.py: MODEL_1 -> MODEL_0
- activity37: Uses MODEL_0 for classification, MODEL_1 for code feedback

This works with the existing environment variable setup without requiring changes to vars.sh.

activity.py

@@ -91,7 +91,7 @@ def get_activity_content(file_path):
 
 
 def loop_through_steps_until_question(
-    activity_content, activity_state, room_name, username, classifier_model="MODEL_1", feedback_model="MODEL_1"
+    activity_content, activity_state, room_name, username, classifier_model="MODEL_0", feedback_model="MODEL_0"
 ):
     room = get_room(room_name)
 
@@ -223,9 +223,9 @@ def start_activity(room_name, s3_file_path, username):
         db.session.commit()
 
         # Get model configuration from activity content if specified
-        # Default to MODEL_1 (Hermes) for both - fast, accurate, and always available
-        classifier_model = activity_content.get("classifier_model", "MODEL_1")
-        feedback_model = activity_content.get("feedback_model", "MODEL_1")
+        # Default to MODEL_0 (Hermes) for both - fast, accurate, and always available
+        classifier_model = activity_content.get("classifier_model", "MODEL_0")
+        feedback_model = activity_content.get("feedback_model", "MODEL_0")
 
         # Loop through steps until a question is found or the end is reached
         loop_through_steps_until_question(
@@ -335,7 +335,7 @@ def execute_processing_script(metadata, script):
     return local_env["script_result"]
 
 
-def handle_activity_response(room_name, user_response, username, model="MODEL_1"):
+def handle_activity_response(room_name, user_response, username, model="MODEL_0"):
     with app.app_context():
         room = get_room(room_name)
         activity_state = ActivityState.query.filter_by(room_id=room.id).first()
@@ -347,9 +347,9 @@ def handle_activity_response(room_name, user_response, username, model="MODEL_1"
         activity_content = get_activity_content(activity_state.s3_file_path)
 
         # Get activity-level model defaults
-        # Default to MODEL_1 (Hermes) for both - fast, accurate, and always available
-        default_classifier_model = activity_content.get("classifier_model", "MODEL_1")
-        default_feedback_model = activity_content.get("feedback_model", "MODEL_1")
+        # Default to MODEL_0 (Hermes) for both - fast, accurate, and always available
+        default_classifier_model = activity_content.get("classifier_model", "MODEL_0")
+        default_feedback_model = activity_content.get("feedback_model", "MODEL_0")
 
         try:
             # Find the current section and step
@@ -921,7 +921,7 @@ def handle_activity_response(room_name, user_response, username, model="MODEL_1"
             )
 
 
-def display_activity_info(room_name, username, model="MODEL_1"):
+def display_activity_info(room_name, username, model="MODEL_0"):
     with app.app_context():
         room = get_room(room_name)
         activity_state = ActivityState.query.filter_by(room_id=room.id).first()
@@ -1009,7 +1009,7 @@ def display_activity_info(room_name, username, model="MODEL_1"):
             print(f"Exception: {e}")
 
 
-def generate_grading(chat_history, rubric, model="MODEL_1"):
+def generate_grading(chat_history, rubric, model="MODEL_0"):
     # Use provided model or fall back to default
     if model and model != "None":
         openai_client, model_name = get_openai_client_and_model(model)
@@ -1061,7 +1061,7 @@ def get_next_step(activity_content, current_section_id, current_step_id):
 
 
 # Categorize the user's response.
-def categorize_response(question, response, buckets, tokens_for_ai, model="MODEL_1"):
+def categorize_response(question, response, buckets, tokens_for_ai, model="MODEL_0"):
     # Use provided model or fall back to default
     if model and model != "None":
         openai_client, model_name = get_openai_client_and_model(model)
@@ -1142,7 +1142,7 @@ def generate_ai_feedback(
     username,
     json_metadata,
     json_new_metadata,
-    model="MODEL_1",
+    model="MODEL_0",
 ):
     # Use provided model or fall back to default
     if model and model != "None":
@@ -1180,7 +1180,7 @@ def provide_feedback(
     username,
     json_metadata,
     json_new_metadata,
-    model="MODEL_1",
+    model="MODEL_0",
 ):
     feedback = ""
     if "ai_feedback" in transition:
@@ -1211,7 +1211,7 @@ def provide_feedback_prompts(
     json_metadata,
     json_new_metadata,
     legacy_tokens_for_ai="",
-    model="MODEL_1",
+    model="MODEL_0",
 ):
     """Generate feedback from multiple prompts"""
     feedback_messages = []
@@ -1331,7 +1331,7 @@ def provide_feedback_prompts(
     return feedback_messages
 
 
-def translate_text(text, target_language, model="MODEL_1"):
+def translate_text(text, target_language, model="MODEL_0"):
     # Guard clause for default language
     target_language = target_language.lower().split()
 

research/activity37-programming-languages.yaml

@@ -2,25 +2,13 @@ default_max_attempts_per_step: 3
 
 # Model configuration
 # classifier_model: Fast classification into buckets (correct, partial, etc.)
-#   MODEL_1 = Hermes-3-Llama-3.1-8B (always available, great for role-play)
+#   MODEL_0 = Hermes (your stable default)
 #
 # feedback_model: Code generation and feedback
-#   MODEL_3 = Qwen3-Coder-30B-A3B-Instruct (specialized for code)
-#   Recommended: hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M
+#   MODEL_1 = Qwen (specialized for code in your setup)
 #
-#   Setup with llama.cpp:
-#     1. Download: huggingface-cli download unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF \
-#                    Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf
-#     2. Run: llama-server -m Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf \
-#               --host 0.0.0.0 --port 8080 -ngl 99
-#     3. Set env: export MODEL_ENDPOINT_3=http://localhost:8080/v1
-#                 export MODEL_API_KEY_3=dummy
-#
-#   Or use with ollama:
-#     ollama run unsloth/qwen3-coder:30b-instruct-q4_K_M
-#
-classifier_model: "MODEL_1"
-feedback_model: "MODEL_3"
+classifier_model: "MODEL_0"
+feedback_model: "MODEL_1"
 
 tokens_for_ai_rubric: |
   Evaluate the student's understanding of programming concepts in their chosen language.