Update app.py

app.py

@@ -1,5 +1,16 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
+import torch
+from transformers import AutoTokenizer, Gemma3ForCausalLM
+
+model_path = "SRP-base-model-training/gemma_3_800M_sft_v2_translation-kazparc_latest"
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = Gemma3ForCausalLM.from_pretrained(
+    model_path,
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto"
+)
+model.eval()
 
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
@@ -7,62 +18,65 @@ For more information on `huggingface_hub` Inference API support, please check th
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
+def respond(message, history, system_message, max_tokens, temperature, top_p):
     messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
+    
+    # Rebuild full chat history
+    for user_msg, assistant_msg in history:
+        if user_msg:
+            messages.append({"role": "user", "content": user_msg})
+        if assistant_msg:
+            messages.append({"role": "assistant", "content": assistant_msg})
     messages.append({"role": "user", "content": message})
 
-    response = ""
+    # Convert chat to single prompt
+    prompt = ""
+    for msg in messages:
+        role = msg["role"]
+        content = msg["content"]
+        if role == "system":
+            prompt += f"[SYSTEM] {content}\n"
+        elif role == "user":
+            prompt += f"[USER] {content}\n"
+        elif role == "assistant":
+            prompt += f"[ASSISTANT] {content}\n"
+    prompt += "[ASSISTANT]"
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    # Tokenize
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    input_len = inputs["input_ids"].shape[-1]
 
-        response += token
-        yield response
+    # Generate tokens (with streaming behavior)
+    generated_text = ""
+    with torch.no_grad():
+        output_ids = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            do_sample=True,
+            temperature=temperature,
+            top_p=top_p,
+            repetition_penalty=1.2,
+            pad_token_id=tokenizer.eos_token_id
+        )
+        output = output_ids[0][input_len:]
+        for i in range(output.shape[0]):
+            token = output[i].unsqueeze(0)
+            text_piece = tokenizer.decode(token, skip_special_tokens=True)
+            generated_text += text_piece
+            yield generated_text
 
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
+# Gradio UI
 demo = gr.ChatInterface(
     respond,
-    additional_inputs_accordion=gr.Accordion(
-        label="⚙️ Parameters", open=False, render=False
-    ),
+    additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False),
     additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Textbox(value="You are a helpful assistant.", label="System message"),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
     ],
     theme="Ocean",
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()