Update app.py

app.py

@@ -1,59 +1,65 @@
 import gradio as gr
-from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
+from transformers import AutoTokenizer, AutoModelForCausalLM
 import torch
-import soundfile as sf
-import numpy as np
 
-# Device configuration
+# Device
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Supported languages (for dropdown, Roman transliteration recommended for non-English)
-languages = ["English", "Hindi", "Urdu", "Arabic", "Turkish", "Persian", "Malay", 
-             "Swahili", "Bengali", "Hausa"]
-
-# Load model & processor once
-model_name = "microsoft/speecht5_tts"
-processor = SpeechT5Processor.from_pretrained(model_name)
-model = SpeechT5ForTextToSpeech.from_pretrained(model_name).to(device)
-vocoder = SpeechT5HifiGan.from_pretrained(model_name).to(device)
-
-# Fixed random speaker embedding for demo
-speaker_embedding = torch.randn(1, 512).to(device)
-
-# Text-to-Speech function
-def text_to_speech(text, language):
-    """
-    Convert text to speech using SpeechT5 model.
-    For non-English languages, Roman transliteration is recommended.
-    """
-    inputs = processor(text=text, return_tensors="pt").to(device)
-    
-    with torch.no_grad():
-        # Generate mel-spectrogram
-        speech = model.generate_speech(inputs["input_ids"], speaker_embeddings=speaker_embedding)
-        # Convert mel to waveform
-        audio_waveform = vocoder(speech.squeeze(0))
-    
-    # Convert to 1D numpy float32 for Gradio compatibility
-    audio_np = audio_waveform.squeeze().cpu().numpy().astype(np.float32)
-    samplerate = processor.feature_extractor.sampling_rate
-    
-    # Optional: save output
-    sf.write("output.wav", audio_np, samplerate)
-    
-    return (audio_np, samplerate)
-
-# Gradio Interface
-iface = gr.Interface(
-    fn=text_to_speech,
-    inputs=[
-        gr.Textbox(lines=3, placeholder="Type your text here..."),
-        gr.Dropdown(languages, label="Select Language")
-    ],
-    outputs=gr.Audio(type="numpy", autoplay=True),
-    title="Multi-Language TTS (SpeechT5)",
-    description="Type text and select language. Roman transliteration recommended for non-English languages."
+# Load Model
+model_name = "Hyprlyf/hypr1-instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto"
 )
 
-# Launch
-iface.launch(share=True)
+# Chat function
+def chat_with_model(user_input, history=[]):
+    # Combine history into context
+    context = ""
+    for h in history:
+        context += f"User: {h[0]}\nAssistant: {h[1]}\n"
+    context += f"User: {user_input}\nAssistant:"
+
+    inputs = tokenizer(context, return_tensors="pt").to(device)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=256,
+            temperature=0.7,
+            top_p=0.9,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id
+        )
+
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    # Extract only assistant's last reply
+    if "Assistant:" in response:
+        reply = response.split("Assistant:")[-1].strip()
+    else:
+        reply = response.strip()
+
+    history.append((user_input, reply))
+    return history, history
+
+# Gradio Chatbot UI
+with gr.Blocks() as demo:
+    gr.Markdown("# 🤖 Hyprlyf/hypr1-instruct Chatbot")
+
+    chatbot = gr.Chatbot()
+    msg = gr.Textbox(placeholder="Type your message here...")
+    clear = gr.Button("Clear")
+
+    state = gr.State([])
+
+    def respond(message, state):
+        state, updated_history = chat_with_model(message, state)
+        return updated_history, state
+
+    msg.submit(respond, [msg, state], [chatbot, state])
+    clear.click(lambda: ([], []), None, [chatbot, state])
+
+demo.launch(share=True)