Added gradio_client API

.gitignore

@@ -9,3 +9,4 @@ build/
 .DS_Store
 *.zip
 *.wav
+.venv/

app.py

@@ -2,38 +2,23 @@ import gradio as gr
 import spaces
 import os, torch, io
 import json
-
-os.system("python -m unidic download")
+import re
+# os.system("python -m unidic download")
 import httpx
-
 # print("Make sure you've downloaded unidic (python -m unidic download) for this WebUI to work.")
 from melo.api import TTS
 import tempfile
 import wave
 from pydub import AudioSegment
-from transformers import (
-    AutoModelForCausalLM,
-    AutoTokenizer,
-    BitsAndBytesConfig,
-)
-
-quantization_config = BitsAndBytesConfig(
-    load_in_4bit=True, bnb_4bit_compute_dtype=torch.float16
-)
-
-model = AutoModelForCausalLM.from_pretrained(
-    "NousResearch/Hermes-2-Pro-Llama-3-8B",
-    quantization_config=quantization_config,
-)
-tok = AutoTokenizer.from_pretrained("NousResearch/Hermes-2-Pro-Llama-3-8B")
-terminators = [tok.eos_token_id, tok.convert_tokens_to_ids("<|eot_id|>")]
-
+from gradio_client import Client
 
+client = Client("eswardivi/AIO_Chat")
 def fetch_text(url):
     print("Entered Webpage Extraction")
     prefix_url = "https://r.jina.ai/"
     url = prefix_url + url
-    response = httpx.get(url, timeout=60.0)
+    response = httpx.get(url, timeout=120.0)
+    print("Response Received")
     return response.text
 
 
@@ -51,23 +36,19 @@ def synthesize(article_url, progress=gr.Progress()):
         }
         """
 
-    chat = [
-        {
-            "role": "user",
-            "content": f"{text} \n Convert the text as Elaborate Conversation between two people as Podcast.\nfollowing this template \n {template}",
-        }
-    ]
-    messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
-    model_inputs = tok([messages], return_tensors="pt").to(device)
-
-    text = model.generate(
-        model_inputs,
-        max_new_tokens=1024,
-        do_sample=True,
-        temperature=0.9,
-        eos_token_id=terminators,
+    result = client.predict(
+		f"{text} \n Convert the text as Elaborate Conversation between two people as Podcast.\nfollowing this template and return only JSON \n {template}",
+		0.9,	
+		True,	
+		1024,
+		api_name="/chat"
     )
-
+    pattern = r"\{(?:[^{}]|(?:\{[^{}]*\}))*\}"
+    json_match = re.search(pattern, result)
+    if json_match:
+        conversation=json_match.group()
+    else:
+        conversation = template
     speed = 1.0
     models = {
         "EN": TTS(language="EN", device=device),
@@ -75,7 +56,7 @@ def synthesize(article_url, progress=gr.Progress()):
     speakers = ["EN-Default", "EN-US"]
 
     combined_audio = AudioSegment.empty()
-    conversation = json.loads(text)
+    conversation = json.loads(conversation)
     for i, turn in enumerate(conversation["conversation"]):
         bio = io.BytesIO()
         text = turn["text"]
@@ -103,4 +84,4 @@ with gr.Blocks() as demo:
     aud = gr.Audio(interactive=False)
     btn.click(synthesize, inputs=[text], outputs=[aud])
 
-demo.queue(api_open=True, default_concurrency_limit=10).launch(show_api=True)
+demo.queue(api_open=True, default_concurrency_limit=10).launch(show_api=True,share=True)

requirements.txt

@@ -1,5 +1,6 @@
 gradio
 spaces
+gradio_client
 tqdm
 #torch
 torchaudio