change code of app to match the MC GHub

app.py

@@ -1,53 +1,71 @@
-import gradio as gr
-import argparse
+from transformers import AutoTokenizer
+import transformers
 import os
-import json
-from vllm import LLM, SamplingParams
+import sys
+import fire
+import torch
+import gradio as gr
 
 
-def parse_args():
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--base_model", type=str)  # model path
-    parser.add_argument("--n_gpus", type=int, default=1)  # n_gpu
-    return parser.parse_args()
+def main(
+    base_model="ise-uiuc/Magicoder-S-DS-6.7B",
+    device="cuda:0",
+    port=8080,
+):
+    tokenizer = AutoTokenizer.from_pretrained(base_model)
+    pipeline = transformers.pipeline(
+        "text-generation",
+        model=base_model,
+        torch_dtype=torch.float16,
+        device=device
+    )
+    def evaluate_magicoder(
+        instruction,
+        temperature=1,
+        max_new_tokens=2048,
+    ):
+        MAGICODER_PROMPT = """You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.
 
-def predict(message, history, system_prompt, temperature, max_tokens):
-    instruction = "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. "
-    for human, assistant in history:
-        instruction += 'USER: '+ human + ' ASSISTANT: '+ assistant + '</s>'
-    instruction += 'USER: '+ message + ' ASSISTANT:'
-    problem = [instruction]
-    stop_tokens = ["USER:", "USER", "ASSISTANT:", "ASSISTANT"]
-    sampling_params = SamplingParams(temperature=temperature, top_p=1, max_tokens=max_tokens, stop=stop_tokens)
-    completions = llm.generate(problem, sampling_params)
-    for output in completions:
-        prompt = output.prompt
-        print('==========================question=============================')
-        print(prompt)
-        generated_text = output.outputs[0].text
-        print('===========================answer=============================')
-        print(generated_text)
-        for idx in range(len(generated_text)):
-                yield generated_text[:idx+1]
+@@ Instruction
+{instruction}
 
+@@ Response
+""" 
+        prompt = MAGICODER_PROMPT.format(instruction=instruction)
 
-if __name__ == "__main__":
-    args = parse_args()
-    llm = LLM(model=args.base_model, tensor_parallel_size=args.n_gpus)
-    gr.ChatInterface(
-        predict,
-        title="LLM playground - WizardLM-13B-V1.2",
-        description="This is a LLM playground for WizardLM-13B-V1.2, github: https://github.com/nlpxucan/WizardLM, huggingface: https://huggingface.co/WizardLM",
-        theme="soft",
-        chatbot=gr.Chatbot(height=1400, label="Chat History",),
-        textbox=gr.Textbox(placeholder="input", container=False, scale=7),
-        retry_btn=None,
-        undo_btn="Delete Previous",
-        clear_btn="Clear",
-        additional_inputs=[
-            gr.Textbox("A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.", label="System Prompt"),
-            gr.Slider(0, 1, 0.9, label="Temperature"),
-            gr.Slider(100, 2048, 1024, label="Max Tokens"),
+        sequences = pipeline(
+            prompt,
+            temperature=temperature,
+            max_new_tokens=max_new_tokens,
+        )
+        for seq in sequences:
+            print('==========================question=============================')
+            print(prompt)
+            generated_text = seq['generated_text'].replace(prompt, "")
+            print('===========================answer=============================')
+            print(generated_text)
+            return generated_text
+
+    gr.Interface(
+        fn=evaluate_magicoder,
+        inputs=[
+            gr.components.Textbox(
+                lines=3, label="Instruction", placeholder="Anything you want to ask Magicoder ?"
+            ),
+            gr.components.Slider(minimum=0, maximum=1, value=0, label="Temperature"),
+            gr.components.Slider(
+                minimum=1, maximum=2048, step=1, value=128, label="Max tokens"
+            ),
         ],
-        additional_inputs_accordion_name="Parameters",
-    ).queue().launch(share=False, server_port=7870)
+        outputs=[
+            gr.components.Textbox(
+                lines=30,
+                label="Output",
+            )
+        ],
+        title="Magicoder",
+        description="This is a LLM playground for Magicoder! Follow us on Github: https://github.com/ise-uiuc/magicoder and Huggingface: https://huggingface.co/ise-uiuc."
+    ).queue().launch(share=True, server_port=port)
+
+if __name__ == "__main__":
+    fire.Fire(main)

requirements.txt

@@ -1 +1 @@
-vllm
+fire