Update app.py

app.py

@@ -9,14 +9,19 @@ from safetensors.torch import load_file as load_safetensors
 # 🔧 Model versions configuration
 # ----------------------------
 MODEL_VERSIONS = {
-        "Beeper v3 (Multi-Concept)": {
+    "Beeper v4 (Advanced)": {
+        "repo_id": "AbstractPhil/beeper-rose-v4",
+        "model_file": "beeper_rose_final.safetensors",
+        "description": "Beeper v4 with nearly 40% the full corpus training - the most capable version currently."
+    },
+    "Beeper v3 (Multi-Concept)": {
         "repo_id": "AbstractPhil/beeper-rose-v3",
         "model_file": "beeper_rose_final.safetensors",
-        "description": "Beeper v3 with 30+ epochs including reasoning, math, coding, and more."
+        "description": "Beeper v3 with 30+ epochs including reasoning, math, and ethics"
     },
     "Beeper v2 (Extended)": {
         "repo_id": "AbstractPhil/beeper-rose-v2",
-        "model_file": "beeper_rose_final.safetensors",
+        "model_file": "beeper_final.safetensors",
         "description": "Beeper v2 with extended training (~15 epochs)"
     },
     "Beeper v1 (Original)": {
@@ -90,14 +95,21 @@ def load_model_version(version_name):
     except Exception as e:
         return f"Error loading {version_name}: {str(e)}"
 
-# Load default model on startup
-load_status = load_model_version("Beeper v3 (Multi-Concept)")
+# Load default model on startup - try v4 first, fallback to v3
+try:
+    load_status = load_model_version("Beeper v4 (Advanced)")
+    if "Error" in load_status:
+        print(f"v4 not ready yet: {load_status}")
+        load_status = load_model_version("Beeper v3 (Multi-Concept)")
+except:
+    load_status = load_model_version("Beeper v3 (Multi-Concept)")
+    
 print(load_status)
 
 # ----------------------------
 # 💬 Gradio Chat Wrapper
 # ----------------------------
-def beeper_reply(message, history, model_version, temperature=None, top_k=None, top_p=None):
+def beeper_reply(message, history, model_version, temperature=None, top_k=None, top_p=None, max_new_tokens=80):
     global infer, tok, current_version
     
     # Load model if version changed
@@ -135,11 +147,11 @@ def beeper_reply(message, history, model_version, temperature=None, top_k=None,
         tok=tok,
         cfg=config,
         prompt=prompt,
-        max_new_tokens=80,  # Shorter to avoid rambling
-        temperature=float(temperature) * 0.8,  # Slightly lower temp
+        max_new_tokens=max_new_tokens,  # Shorter to avoid rambling
+        temperature=float(temperature),  # Slightly lower temp
         top_k=int(top_k),
         top_p=float(top_p),
-        repetition_penalty=1.2,  # Higher penalty for repetition
+        repetition_penalty=1.1,  # Higher penalty for repetition
         presence_penalty=0.8,    # Higher presence penalty
         frequency_penalty=0.1,    # Add frequency penalty
         device=device,
@@ -211,12 +223,12 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         with gr.Column(scale=3):
             model_dropdown = gr.Dropdown(
                 choices=list(MODEL_VERSIONS.keys()),
-                value="Beeper v3 (Multi-Concept)",
+                value="Beeper v3 (Multi-Concept)",  # Default to v3 since v4 might not be ready
                 label="Select Beeper Version",
                 info="Choose which version of Beeper to chat with"
             )
         with gr.Column(scale=7):
-            version_info = gr.Markdown("**Current:** Beeper v3 with 30+ epochs including reasoning, math, coding, and more.")
+            version_info = gr.Markdown("**Current:** Beeper v3 with 30+ epochs including reasoning, math, and ethics")
     
     # Update version info when dropdown changes
     def update_version_info(version_name):
@@ -231,7 +243,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     
     # Chat interface
     chatbot = gr.Chatbot(label="Chat with Beeper", type="tuples", height=400)
-    msg = gr.Textbox(label="Message", placeholder="Type your message here... She will probably complete it for now, but maybe she'll answer.")
+    msg = gr.Textbox(label="Message", placeholder="Type your message here...")
     
     with gr.Row():
         with gr.Column(scale=2):
@@ -240,6 +252,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             top_k_slider = gr.Slider(1, 100, value=40, step=1, label="Top-k")
         with gr.Column(scale=2):
             top_p_slider = gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-p")
+        with gr.Column(scale=2):
+            max_new_tokens_slider = gr.Slider(20, 512, value=128, step=1, label="Max-new-tokens")
     
     with gr.Row():
         submit = gr.Button("Send", variant="primary")
@@ -261,18 +275,18 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     def respond(message, chat_history, model_version, temperature, top_k, top_p):
         if not chat_history:
             chat_history = []
-        response = beeper_reply(message, chat_history, model_version, temperature, top_k, top_p)
+        response = beeper_reply(message, chat_history, model_version, temperature, top_k, top_p, max_new_tokens_slider)
         chat_history.append([message, response])
         return "", chat_history
     
     msg.submit(
         respond, 
-        [msg, chatbot, model_dropdown, temperature_slider, top_k_slider, top_p_slider], 
+        [msg, chatbot, model_dropdown, temperature_slider, top_k_slider, top_p_slider, max_new_tokens_slider], 
         [msg, chatbot]
     )
     submit.click(
         respond, 
-        [msg, chatbot, model_dropdown, temperature_slider, top_k_slider, top_p_slider], 
+        [msg, chatbot, model_dropdown, temperature_slider, top_k_slider, top_p_slider, max_new_tokens_slider], 
         [msg, chatbot]
     )
     clear.click(lambda: None, None, chatbot, queue=False)