add max_new_tokens to huggingface models; modifed graph initiation;

pages/play.py

@@ -9,7 +9,7 @@ from support.game_settings import TEAM_MODEL_PRESETS
 from support.utils import format_messages_as_feed, generate_team_html, plot_game_board_with_guesses
 
 
-graph = MyGraph()
+# graph = MyGraph()
 TEAM_OPTIONS = list(TEAM_MODEL_PRESETS.keys()) + ["random"]
 
 
@@ -54,6 +54,7 @@ with gr.Blocks(fill_width=True) as demo:
     turn_state = gr.State(value=0)
     winners_state = gr.State(value=None)
     api_keys_state = gr.State(value={})
+    graph_instance = gr.State() 
 
     # Team selection (visible initially)
     with gr.Row(elem_id="game_setup_row"):
@@ -222,7 +223,8 @@ with gr.Blocks(fill_width=True) as demo:
                 game_started: False,
                 players_state: [],
                 board_state: {},
-                messages_state: []
+                messages_state: [],
+                graph_instance: None  # ADD THIS
             }
 
         # Create game with validated API keys
@@ -231,6 +233,9 @@ with gr.Blocks(fill_width=True) as demo:
         html = generate_team_html(game.players, starting_team)
         board_img = plot_game_board_with_guesses(game.board)
 
+        # CREATE GRAPH INSTANCE FOR THIS SESSION
+        graph = MyGraph()  # ADD THIS
+
         return {
             game_state: game,
             new_game_btn: gr.update(visible=True),
@@ -243,7 +248,8 @@ with gr.Blocks(fill_width=True) as demo:
             board_state: game.board,
             messages_state: [],
             api_keys_state: api_keys,
-            validation_status: gr.update(visible=True, value="✅ Game started successfully!")
+            validation_status: gr.update(visible=True, value="✅ Game started successfully!"),
+            graph_instance: graph  # ADD THIS
         }
 
     def generate_new_game(red_team_choice, blue_team_choice, openai_key, google_key, anthropic_key, hf_key, current_api_keys):
@@ -262,7 +268,10 @@ with gr.Blocks(fill_width=True) as demo:
         html = generate_team_html(game.players, starting_team)
         board_img = plot_game_board_with_guesses(game.board)
 
-        return game, html, board_img, game.players, game.board, [], gr.update(visible=True), api_keys
+        # CREATE NEW GRAPH INSTANCE
+        graph = MyGraph()  # ADD THIS
+
+        return game, html, board_img, game.players, game.board, [], gr.update(visible=True), api_keys, graph  # ADD graph to return
 
     def show_red_boss_input():
         return gr.Column(visible=True)
@@ -310,7 +319,7 @@ with gr.Blocks(fill_width=True) as demo:
         is_human_playing = True
         return html, players, gr.Column(visible=False), gr.Row(visible=False), gr.Row(visible=True), is_human_playing, gr.Button(visible=False)
 
-    async def process_message(user_msg, messages, players, board, dropdown, guessed, chat_history, is_human_playing, turn):
+    async def process_message(user_msg, messages, players, board, dropdown, guessed, chat_history, is_human_playing, turn, graph):  # ADD graph parameter
         """Process user message and stream responses"""
 
         accumulated_messages = list(messages) if messages else []
@@ -331,10 +340,10 @@ with gr.Blocks(fill_width=True) as demo:
         # Don't update if no game is active
         if game_state is None:
             return gr.update()  # No update
-        
+
         if not guessed_words or not board:
             return gr.update()  # No update
-            
+
         logger.info(f"Calling update_plot: {guessed_words}")
         board_img = plot_game_board_with_guesses(board, guessed_words)
         return board_img
@@ -385,10 +394,10 @@ with gr.Blocks(fill_width=True) as demo:
     start_btn.click(
         fn=start_game,
         inputs=[red_team_dropdown, blue_team_dropdown, openai_api_input, google_api_input, anthropic_api_input, hf_api_input],
-        outputs=[game_state, new_game_btn, start_btn, game_content, team_html, board_plot, game_started, players_state, board_state, messages_state, api_keys_state, validation_status]
+        outputs=[game_state, new_game_btn, start_btn, game_content, team_html, board_plot, game_started, players_state, board_state, messages_state, api_keys_state, validation_status, graph_instance]  # ADD graph_instance
     )
 
-    # Update new game button click handler
+    # Update new_game_btn.click
     new_game_btn.click(
         fn=generate_new_game,
         inputs=[
@@ -408,7 +417,8 @@ with gr.Blocks(fill_width=True) as demo:
             board_state,
             messages_state,
             play_as_boss_section,
-            api_keys_state
+            api_keys_state,
+            graph_instance  # ADD THIS
         ]
     )
 
@@ -464,12 +474,13 @@ with gr.Blocks(fill_width=True) as demo:
         outputs=[blue_boss_name_input]
     )
 
+    # Update send_btn.click - THIS IS YOUR MAIN QUESTION
     send_btn.click(
         fn=lambda: gr.Button(interactive=False),
         outputs=[send_btn]
     ).then(
         fn=process_message,
-        inputs=[user_input, messages_state, players_state, board_state, dropdown, guessed_words_state, chat_history_state, is_human_playing_state, turn_state],
+        inputs=[user_input, messages_state, players_state, board_state, dropdown, guessed_words_state, chat_history_state, is_human_playing_state, turn_state, graph_instance],  # ADD graph_instance
         outputs=[message_feed, guessed_words_state, messages_state, board_state, chat_history_state, winners_state]
     ).then(
         fn=lambda: "",

support/game_settings.py

@@ -482,7 +482,7 @@ TEAM_MODEL_PRESETS = {
         "players": ["claude-3-5-haiku-20241022", "claude-3-haiku-20240307"]
     },
     "opensource": {
-        "boss": "deepseek-ai/DeepSeek-V3.1",
+        "boss": "moonshotai/Kimi-K2-Thinking",  # "deepseek-ai/DeepSeek-V3.1",
         "captain": "deepseek-ai/DeepSeek-R1",  # "Qwen/Qwen3-235B-A22B-Instruct-2507",
         "players": ["openai/gpt-oss-120b", "openai/gpt-oss-20b"]
     },

support/load_models.py

@@ -86,7 +86,8 @@ def create_model(model_name: str, api_keys: dict):
             streaming=True,
             temperature=0,
             top_p=0,
-            huggingfacehub_api_token=api_keys['HUGGINGFACEHUB_API_TOKEN']
+            huggingfacehub_api_token=api_keys['HUGGINGFACEHUB_API_TOKEN'],
+            max_new_tokens=1600
         )
 
         return ChatHuggingFace(llm=llm, verbose=True)