puffy310/ZeroGPU-DeepSeek-V2-LiteCoder

1import gradio as gr2import spaces3from transformers import AutoTokenizer, AutoModelForCausalLM4import torch5 6tokenizer = AutoTokenizer.from_pretrained("BirdL/DeepSeek-Coder-V2-Lite-Instruct-FlashAttnPatch", trust_remote_code=True)7model = AutoModelForCausalLM.from_pretrained("BirdL/DeepSeek-Coder-V2-Lite-Instruct-FlashAttnPatch", trust_remote_code=True, device_map="auto", torch_dtype=torch.float16)    8 9@spaces.GPU10def respond(message, history):11    inputs = tokenizer(message, return_tensors="pt").input_ids.to("cuda")12    outputs = model.generate(inputs, max_new_tokens=224, do_sample=False, top_k=50, top_p=0.95, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)13    ouputs = (tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True))14    return outputs15 16demo = gr.ChatInterface(respond)17 18if __name__ == "__main__":19    demo.launch()