using openai whisper + spaces api for chat

libs/transformer/get_chat_gradio.py

@@ -0,0 +1,14 @@
+from gradio_client import Client
+
+def get_chat_gradio(text: str):
+    client = Client("Xkev/Llama-3.2V-11B-cot")
+    result = client.predict(
+            message={"text":text,"files":[]},
+            max_new_tokens=512,
+            api_name="/chat"
+    )
+    return result
+
+# res = get_chat_gradio("Hello, let's chat with me")
+
+# print(res)

libs/transformer/open_ai_whisper.py

@@ -2,27 +2,36 @@ import torch
 from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
 
 
-device = "cuda:0" if torch.cuda.is_available() else "cpu"
-torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
-
-model_id = "openai/whisper-large-v3-turbo"
-
-model = AutoModelForSpeechSeq2Seq.from_pretrained(
-    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
-)
-model.to(device)
-
-processor = AutoProcessor.from_pretrained(model_id)
-
-pipe = pipeline(
-    "automatic-speech-recognition",
-    model=model,
-    tokenizer=processor.tokenizer,
-    feature_extractor=processor.feature_extractor,
-    torch_dtype=torch_dtype,
-    device=device,
-    return_timestamps=True
-)
-
-result = pipe("https://static.langkingdom.com/user_playlist_practice_videos/bdfd406cb3c62603f653fa02d93fcae8.mov")
-print(result["text"])
+def open_ai_whisper_api(url: str):
+    device = "cuda:0" if torch.cuda.is_available() else "cpu"
+    torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+
+    model_id = "openai/whisper-large-v3-turbo"
+
+    model = AutoModelForSpeechSeq2Seq.from_pretrained(
+        model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
+    )
+    model.to(device)
+
+    processor = AutoProcessor.from_pretrained(model_id)
+
+    pipe = pipeline(
+        "automatic-speech-recognition",
+        model=model,
+        tokenizer=processor.tokenizer,
+        feature_extractor=processor.feature_extractor,
+        torch_dtype=torch_dtype,
+        device=device,
+        return_timestamps=True,
+    )
+
+    generate_kwargs = {
+    "return_timestamps": True,
+    "language": "english"
+    }
+
+
+    result = pipe(url, generate_kwargs=generate_kwargs)
+    print('vao day')
+    return result.get('text'), result.get('chunks')
+

routers/get_chatrespone.py

@@ -11,6 +11,8 @@ from pydantic import BaseModel
 from fastapi.responses import StreamingResponse
 from langchain_ollama import ChatOllama, OllamaLLM
 
+from libs.transformer.get_chat_gradio import get_chat_gradio
+
 load_dotenv()
 HUGGINGFACEHUB_API_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN", )
 
@@ -29,18 +31,20 @@ async def get_chat_respone(body: ChatInputForm, api_key: str = Depends(get_api_k
     prompt = get_prompt(body.prompt)
 
     try:
-        llm = OllamaLLM(
-            model=body.repo_id,
-            temperature=0.2,
-            # huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN,
-        )
-
-        messages = [
-            ("system", prompt),
-            ("human", body.textInput)
-        ]
-
-        response = llm.stream(messages)
+        # llm = OllamaLLM(
+        #     model=body.repo_id,
+        #     temperature=0.2,
+        #     # huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN,
+        # )
+
+        # messages = [
+        #     ("system", prompt),
+        #     ("human", body.textInput)
+        # ]
+
+        # response = llm.stream(messages)
+
+        response = get_chat_gradio(body.textInput)
 
         return StreamingResponse(get_response(response), media_type='text/event-stream')
     except Exception:

routers/get_transcript_gradio.py

@@ -6,6 +6,7 @@ from fastapi import APIRouter, Depends, HTTPException, status
 
 from libs.convert_to_audio import convert_to_audio
 from libs.header_api_auth import get_api_key
+from libs.transformer.open_ai_whisper import open_ai_whisper_api
 from libs.transformer.get_transcript_gradio_api import api_gradio_transcribe
 
 
@@ -18,27 +19,36 @@ def get_transcript(audio_path: str, model_size: str = "distil-whisper/distil-sma
 
     output_audio_folder = f"./cached/audio"
 
-    if not os.path.exists(output_audio_folder):
-        os.makedirs(output_audio_folder)
+    # if not os.path.exists(output_audio_folder):
+    #     os.makedirs(output_audio_folder)
 
 
-    output_file = f"{output_audio_folder}/{audio_path.split('/')[-1].split(".")[0]}.mp3"
-    convert_to_audio(audio_path.strip(), output_file)
+    # output_file = f"{output_audio_folder}/{audio_path.split('/')[-1].split(".")[0]}.mp3"
+    # convert_to_audio(audio_path.strip(), output_file)
 
     try:
-        text = api_gradio_transcribe(output_file)
+        text, chunks = open_ai_whisper_api(audio_path)
 
     except Exception as error:
         raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail=f"error>>>: {error}")
-    finally:
-         if os.path.exists(output_file):
-            os.remove(output_file)
+    # finally:
+        #  if os.path.exists(output_file):
+        #     os.remove(output_file)
+
+    listSentences = []
+
+    for chunk in chunks:
+        listSentences.append({
+            "start_time": chunk.get("timestamp")[0],
+            "end_time": chunk.get("timestamp")[1],
+            "text": chunk.get("text")
+        })
 
     et = time.time()
 
     elapsed_time = et - st
 
     return {"text": text,
-            'list_sentence':  [],
+            'list_sentence':  listSentences,
             'elapsed_time': round(elapsed_time, 2)
             }