Update app.py

app.py

@@ -13,9 +13,9 @@ import spaces  # ZeroGPU decorator
 # Using Qwen3-Embedding-4B for 2560 native dimensions (truncate to 1536 for production)
 # Qwen3-4B is optimal for 1536 dims: 60% retention (vs 42.9% for GTE-Qwen2-7B)
 MODEL_NAME = "Qwen/Qwen3-Embedding-4B"
-print(f"🔄 Loading model: {MODEL_NAME}")
+print(f"   Loading model: {MODEL_NAME}")
 model = SentenceTransformer(MODEL_NAME, trust_remote_code=True)
-print(f"✅ Model loaded successfully")
+print(f"   Model loaded successfully")
 print(f"   Native Dimensions: {model.get_sentence_embedding_dimension()}")
 print(f"   Max Seq Length: {model.max_seq_length}")
 print(f"   Matryoshka Support: Yes (truncate to any dimension ≤ {model.get_sentence_embedding_dimension()})")
@@ -140,16 +140,16 @@ def calculate_similarity(text1: str, text2: str, use_instruction: bool) -> dict:
 # Create Gradio interface
 with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
-    # 🎲 FairFate Embeddings API
+    # FairFate Embeddings API
 
     **Powered by Qwen3-Embedding-4B** - Advanced Multilingual Embedding Model
 
-    - 🌍 **100+ Languages** (English, Spanish, French, German, Chinese, Japanese, etc.)
-    - 📐 **2560 Native Dimensions** (matryoshka truncation to 1536 for production)
-    - 📚 **32K Context** (massive text support)
-    - ⚡ **Instruction-Aware** (optimized for RPG content)
-    - 🔬 **Matryoshka Support** (flexible 32-2560 dimensions)
-    - 🏆 **Optimal for 1536 dims** (60% dimension retention)
+    - **100+ Languages** (English, Spanish, French, German, Chinese, Japanese, etc.)
+    - **2560 Native Dimensions** (matryoshka truncation to 1536 for production)
+    - **32K Context** (massive text support)
+    - **Instruction-Aware** (optimized for RPG content)
+    - **Matryoshka Support** (flexible 32-2560 dimensions)
+    - **Optimal for 1536 dims** (60% dimension retention)
 
     Perfect for: Product classification, semantic search, recommendations, multilingual matching
     """)
@@ -187,7 +187,7 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
             inputs=[input_text, use_inst, output_dims],
         )
 
-    with gr.Tab("🔍 Similarity Calculator"):
+    with gr.Tab("Similarity Calculator"):
         gr.Markdown("""
         **Comprehensive Similarity Analysis** - Compare two texts using multiple metrics:
 
@@ -203,12 +203,12 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
 
         with gr.Row():
             with gr.Column():
-                text1 = gr.Textbox(
+                text1 = gr.JSON(
                     label="First Text",
                     placeholder="Storm King's Thunder - Giant-themed D&D adventure",
                     lines=3
                 )
-                text2 = gr.Textbox(
+                text2 = gr.JSON(
                     label="Second Text",
                     placeholder="Princes of the Apocalypse - Elemental evil campaign",
                     lines=3
@@ -230,9 +230,9 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
             inputs=[text1, text2, use_inst_sim],
         )
 
-    with gr.Tab("📖 API Documentation"):
+    with gr.Tab("API Documentation"):
         gr.Markdown("""
-        ## 🚀 Quick Start
+        ## Quick Start
 
         ### Python
 
@@ -297,7 +297,7 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
           }'
         ```
 
-        ## 📊 Parameters
+        ## Parameters
 
         | Parameter | Type | Default | Description |
         |-----------|------|---------|-------------|
@@ -305,7 +305,7 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
         | `use_instruction` | boolean | true | Add instruction prefix (improves accuracy) |
         | `output_dimensions` | number | 1536 | Output size (32-3584, production default: 1536) |
 
-        ## 🎯 Use Cases
+        ## Use Cases
 
         - **Product Classification**: Auto-tag by genre, system, theme
         - **Semantic Search**: Find by meaning, not keywords
@@ -313,20 +313,20 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
         - **Duplicate Detection**: Find similar listings
         - **Multilingual Matching**: Cross-language similarity
 
-        ## ⚡ Performance
+        ## Performance
 
         | Batch Size | GPU Throughput | CPU Throughput |
         |------------|----------------|----------------|
         | 1          | ~800/sec       | ~80/sec        |
         | 32         | ~4000/sec      | ~250/sec       |
 
-        ## 🌍 Supported Languages
+        ## Supported Languages
 
         English, Spanish, French, German, Italian, Portuguese, Russian, Polish, Dutch, Czech,
         Chinese, Japanese, Korean, Arabic, Hebrew, Hindi, Thai, Vietnamese, Indonesian,
         Turkish, Swedish, Norwegian, Danish, Finnish, Greek, Romanian, Hungarian, and 80+ more!
 
-        ## 📝 Citation
+        ## Citation
 
         ```bibtex
         @misc{qwen3-embedding-2025,
@@ -338,25 +338,25 @@ with gr.Blocks(title="FairFate Embeddings API - Qwen3", theme=gr.themes.Soft())
         ```
         """)
 
-    with gr.Tab("ℹ️ Model Info"):
+    with gr.Tab("Model Info"):
         gr.Markdown(f"""
         ## Model Details
 
-        - **Model:** {MODEL_NAME}
-        - **Dimensions:** {model.get_sentence_embedding_dimension()}
-        - **Max Sequence Length:** {model.max_seq_length} tokens
-        - **Languages:** 100+
-        - **License:** Apache 2.0
-        - **Normalization:** L2 normalized (ready for cosine similarity)
+            - **Model:** {MODEL_NAME}
+            - **Dimensions:** {model.get_sentence_embedding_dimension()}
+            - **Max Sequence Length:** {model.max_seq_length} tokens
+            - **Languages:** 100+
+            - **License:** Apache 2.0
+            - **Normalization:** L2 normalized (ready for cosine similarity)
 
         ## Advantages
 
-        ✅ **Best Multilingual Performance** - Top tier on MTEB leaderboard
-        ✅ **Massive Context** - 32K tokens (vs 512 for most models)
-        ✅ **Instruction-Aware** - Can customize for specific domains
-        ✅ **Flexible Dimensions** - 32 to 2560 dimensions (matryoshka truncation)
-        ✅ **Code-Switching** - Handles mixed-language text
-        ✅ **Production Optimized** - 60% retention at 1536 dims (best in class)
+            **Best Multilingual Performance** - Top tier on MTEB leaderboard
+            **Massive Context** - 32K tokens (vs 512 for most models)
+            **Instruction-Aware** - Can customize for specific domains
+            **Flexible Dimensions** - 32 to 2560 dimensions (matryoshka truncation)
+            **Code-Switching** - Handles mixed-language text
+            **Production Optimized** - 60% retention at 1536 dims (best in class)
 
         ## Resources