Update app.py

app.py

@@ -16,29 +16,42 @@ def respond(
     temperature,
     top_p,
 ):
-    messages = [{"role": "system", "content": system_message}]
+    torch.set_default_device("cuda")
 
-    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})
-
-    response = ""
+    tokenizer = AutoTokenizer.from_pretrained(
+        "cognitivecomputations/dolphin-2.8-mistral-7b-v02",
+        trust_remote_code=True
+    )
+    model = AutoModelForCausalLM.from_pretrained(
+        "cognitivecomputations/dolphin-2.8-mistral-7b-v02",
+        torch_dtype="auto",
+        load_in_4bit=True,
+        trust_remote_code=True
+    )
+    history_transformer_format = history + [[message, ""]]
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
+    system_prompt = f"<|im_start|>system\n{system_message}.<|im_end|>"
+    messages = system_prompt + "".join(["".join(["\n<|im_start|>user\n" + item[0], "<|im_end|>\n<|im_start|>assistant\n" + item[1]]) for item in history_transformer_format])
+    input_ids = tokenizer([messages], return_tensors="pt").to('cuda')
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+    generate_kwargs = dict(
+        input_ids,
+        streamer=streamer,
+        max_new_tokens=max_tokens,
+        do_sample=True,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
+        top_k=50,
+        temperature=temperature,
+        num_beams=1
+    )
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+    partial_message = ""
+    for new_token in streamer:
+        partial_message += new_token
+        if '<|im_end|>' in partial_message:
+            break
+        yield partial_message
 
 """
 For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface