Update app.py

app.py

@@ -23,12 +23,12 @@ from sentence_transformers import SentenceTransformer
 # Global variables for pipelines and settings.
 TEXT_PIPELINE = None
 COMPARISON_PIPELINE = None
-NUM_EXAMPLES = 50 
+NUM_EXAMPLES = 100  
 
 @spaces.GPU(duration=300)
 def finetune_small_subset():
     """
-    Fine-tunes the custom DeepSeekV3 model on a small subset of the ServiceNow-AI/R1-Distill-SFT dataset.
+    Fine-tunes the custom R1 model on a small subset of the ServiceNow-AI/R1-Distill-SFT dataset.
     Steps:
       1) Loads the model from "wuhp/myr1" (using files from the "myr1" subfolder via trust_remote_code).
       2) Applies 4-bit quantization and prepares for QLoRA training.
@@ -163,7 +163,7 @@ def ensure_pipeline():
 
 def ensure_comparison_pipeline():
     """
-    Loads a reference DeepSeek model pipeline if not already loaded.
+    Loads the official R1 model pipeline if not already loaded.
     """
     global COMPARISON_PIPELINE
     if COMPARISON_PIPELINE is None:
@@ -180,7 +180,7 @@ def ensure_comparison_pipeline():
 @spaces.GPU(duration=120)
 def predict(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
     """
-    Direct generation without retrieval.
+    Direct generation without retrieval using the custom R1 model.
     """
     pipe = ensure_pipeline()
     out = pipe(
@@ -196,7 +196,7 @@ def predict(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
 @spaces.GPU(duration=120)
 def compare_models(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
     """
-    Compare outputs between your custom model and a reference DeepSeek model.
+    Compare outputs between your custom R1 model and the official R1 model.
     """
     local_pipe = ensure_pipeline()
     comp_pipe = ensure_comparison_pipeline()
@@ -299,34 +299,34 @@ def chat_rag(user_input, history, temperature, top_p, min_new_tokens, max_new_to
 
 # Build the Gradio interface.
 with gr.Blocks() as demo:
-    gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo using Custom DeepSeekV3 Model")
+    gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo using Custom R1 Model")
 
     finetune_btn = gr.Button("Finetune 4-bit (QLoRA) on ServiceNow-AI/R1-Distill-SFT subset (up to 5 min)")
     status_box = gr.Textbox(label="Finetune Status")
     finetune_btn.click(fn=finetune_small_subset, outputs=status_box)
 
-    gr.Markdown("## Direct Generation (No Retrieval)")
+    gr.Markdown("## Direct Generation (No Retrieval) using Custom R1")
     prompt_in = gr.Textbox(lines=3, label="Prompt")
     temperature = gr.Slider(0.0, 1.5, step=0.1, value=0.7, label="Temperature")
     top_p = gr.Slider(0.0, 1.0, step=0.05, value=0.9, label="Top-p")
     min_tokens = gr.Slider(1, 2500, value=50, step=10, label="Min New Tokens")
     max_tokens = gr.Slider(1, 2500, value=200, step=50, label="Max New Tokens")
-    output_box = gr.Textbox(label="DeepSeekV3 Output", lines=8)
-    gen_btn = gr.Button("Generate with DeepSeekV3")
+    output_box = gr.Textbox(label="Custom R1 Output", lines=8)
+    gen_btn = gr.Button("Generate with Custom R1")
     gen_btn.click(
         fn=predict,
         inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
         outputs=output_box
     )
 
-    gr.Markdown("## Compare DeepSeekV3 vs Reference DeepSeek")
+    gr.Markdown("## Compare Custom R1 vs Official R1")
     compare_btn = gr.Button("Compare")
-    out_local = gr.Textbox(label="DeepSeekV3 Output", lines=6)
-    out_deepseek = gr.Textbox(label="Reference DeepSeek Output", lines=6)
+    out_custom = gr.Textbox(label="Custom R1 Output", lines=6)
+    out_official = gr.Textbox(label="Official R1 Output", lines=6)
     compare_btn.click(
         fn=compare_models,
         inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
-        outputs=[out_local, out_deepseek]
+        outputs=[out_custom, out_official]
     )
 
     gr.Markdown("## Chat with Retrieval-Augmented Memory")