test

app.py

@@ -1,11 +1,19 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
 """
 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
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+# Инициализация модели и токенизатора
+MODEL_NAME = "yandex/YandexGPT-5-Lite-8B-pretrain"
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, legacy=False)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME,
+    device_map="cuda" if torch.cuda.is_available() else "cpu",
+    torch_dtype="auto",
+)
 
 def respond(
     message,
@@ -15,33 +23,35 @@ def respond(
     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]})
-
-    messages.append({"role": "user", "content": message})
+    # Формируем контекст из истории
+    full_prompt = f"{system_message}\n\n"
+    for user_msg, assistant_msg in history:
+        if user_msg:
+            full_prompt += f"User: {user_msg}\n"
+        if assistant_msg:
+            full_prompt += f"Assistant: {assistant_msg}\n"
+    full_prompt += f"User: {message}\nAssistant:"
 
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
+    # Токенизация и генерация
+    inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
+        do_sample=True,
+        pad_token_id=tokenizer.eos_token_id,
+        stream=True
+    )
+    
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    # Убираем начальный промпт из ответа
+    response = response[len(full_prompt):]
+    yield response
 
 
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docsx/chatinterface
 """
 demo = gr.ChatInterface(
     respond,