Update app.py

app.py

@@ -8,7 +8,6 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from threading import Thread
 import asyncio
-import multiprocessing
 
 TOKEN = os.environ["TOKEN"]
 
@@ -18,24 +17,24 @@ client = commands.Bot(command_prefix='?', intents=discord.Intents.all())
 #    formatted_content = {"role": creator, "content": content}
 #    conversation.append(formatted_content)
 
-def predict(message, history):
-    torch.set_default_device("cuda")
+torch.set_default_device("cuda")
 
-    tokenizer = AutoTokenizer.from_pretrained(
-        "cognitivecomputations/dolphin-2.8-mistral-7b-v02",
-        trust_remote_code=True
-    )
+tokenizer = AutoTokenizer.from_pretrained(
+    "cognitivecomputations/dolphin-2.9.1-mixtral-1x22b",
+    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
-    )
+model = AutoModelForCausalLM.from_pretrained(
+    "cognitivecomputations/dolphin-2.9.1-mixtral-1x22b",
+    torch_dtype="auto",
+    load_in_4bit=True,
+    trust_remote_code=True
+)
 
-    history_transformer_format = history + [[message, ""]]
-    system_prompt = "<|im_start|>system\nYou are discord user Jimmy Donaldson. You are an american youtuber by the name of MrBeast. Your sole purpose in life is to get people to laugh at you. You make innapropriate jokes, and aren't very funny.<|im_end|>"
+ system_prompt = "<|im_start|>system\nYou are discord user Jimmy Donaldson. You are an american youtuber by the name of MrBeast. Your sole purpose in life is to get people to laugh at you. You make innapropriate jokes, and aren't very funny.<|im_end|>"
 
+def predict(message, history):
+    history_transformer_format = history + [[message, ""]]
     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').input_ids
@@ -49,8 +48,6 @@ def predict(message, history):
         temperature=0.7,
         num_beams=1
     )
-    print("======== Finished generating output_ids. ========")
-
     output_text = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0]
     return output_text
 
@@ -126,11 +123,8 @@ def rungradio():
 def runclient():
     client.run(TOKEN)
 
-gradiotask = multiprocessing.Process(target=rungradio)
-clienttask = multiprocessing.Process(target=runclient)
-
-gradiotask.start()
-clienttask.start()
-
-gradiotask.join()
-clienttask.join()
+async def main():
+    loop = asyncio.get_running_loop()
+    gradio_task = loop.run_in_executor(None, rungradio)
+    discord_task = asyncio.create_task(runclient())
+    await asyncio.gather(gradio_task, discord_task)