test deepseek

app.py

@@ -1,8 +1,8 @@
 # app.py
 import gradio as gr # type: ignore
-# import openai # type: ignore
-# import os
+import os
 
+# import openai # type: ignore
 # # openai.api_key = os.getenv("OPENAI_API_KEY")
 # client = openai.OpenAI()
 # def respond(
@@ -45,39 +45,68 @@ import gradio as gr # type: ignore
 #     print("\n")
 #     yield response #chat reply
 
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
-model_name = "deepseek-ai/deepseek-math-7b-base"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
-# model.generation_config = GenerationConfig.from_pretrained(model_name)
-# model.generation_config.pad_token_id = model.generation_config.eos_token_id
-def deepseek(
-            message, 
-            history: list[tuple[str, str]], 
-            system_message, 
-            max_tokens, 
-            temperature, 
-            top_p):
+# import torch
+# from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
+# model_name = "deepseek-ai/deepseek-math-7b-base"
+# tokenizer = AutoTokenizer.from_pretrained(model_name)
+# model = AutoModelForCausalLM.from_pretrained(model_name)
+# # model.generation_config = GenerationConfig.from_pretrained(model_name)
+# # model.generation_config.pad_token_id = model.generation_config.eos_token_id
+# def deepseek(
+#             message, 
+#             history: list[tuple[str, str]], 
+#             system_message, 
+#             max_tokens, 
+#             temperature, 
+#             top_p):
     
     
-    # messages = [
-    #     {"role": "user", "content": "what is the integral of x^2 from 0 to 2?\nPlease reason step by step, and put your final answer within \\boxed{}."}
-    # ]
-    messages = [
-        {"role": "user", "content": message}
-    ]
+#     # messages = [
+#     #     {"role": "user", "content": "what is the integral of x^2 from 0 to 2?\nPlease reason step by step, and put your final answer within \\boxed{}."}
+#     # ]
+#     messages = [
+#         {"role": "user", "content": message}
+#     ]
+
+#     input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
+#     outputs = model.generate(input_tensor.to(model.device), max_new_tokens=100)
+#     print(outputs)
+#     print("\n")
+#     result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
+#     print(result)
+#     return result
+
+import replicate
+def deepseek_api_replicate(user_message, system_message, max_new_tokens, temperature, top_p):
+    """
+    Gọi DeepSeek Math trên Replicate và trả ngay kết quả.
+
+    Trả về:
+        str hoặc [bytes]: output model sinh ra
+    """
+    # 1. Khởi tạo client và xác thực
+    token = os.getenv("REPLICATE_API_TOKEN")
+    if not token:
+        raise RuntimeError("Missing REPLICATE_API_TOKEN")  # bảo mật bằng biến môi trường
+    client = replicate.Client(api_token=token)          
+
+    # 2. Gọi model
+    output = client.run(
+        "deepseek-ai/deepseek-math-7b-instruct:latest",
+        input={
+            "system_prompt": system_message,
+            "user_prompt":   user_message,
+            "max_new_tokens": max_new_tokens,
+            "temperature":    temperature,
+            "top_p":          top_p
+        }
+    )
 
-    input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
-    outputs = model.generate(input_tensor.to(model.device), max_new_tokens=100)
-    print(outputs)
-    print("\n")
-    result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
-    print(result)
-    return result
+    # 3. Trả kết quả
+    return output
 
 chat = gr.ChatInterface(
-    deepseek, #chat
+    deepseek_api_replicate, #chat
     title="Trợ lý Học Tập AI",
     description="Nhập câu hỏi của bạn về Toán, Lý, Hóa, Văn… và nhận giải đáp chi tiết ngay lập tức!",
     additional_inputs=[

requirements.txt

@@ -1,5 +1,6 @@
 gradio
 # openai
-torch
-transformers
-accelerate
+replicate
+# torch
+# transformers
+# accelerate