Spaces:
Running
Running
| import gradio as gr | |
| import os | |
| os.system('CMAKE_ARGS="-DLLAMA_OPENBLAS=on" FORCE_CMAKE=1 pip install llama-cpp-python') | |
| import wget | |
| from llama_cpp import Llama | |
| import random | |
| import os | |
| import multiprocessing | |
| def get_num_cores(): | |
| """Get the number of CPU cores.""" | |
| return os.cpu_count() | |
| def get_num_threads(): | |
| """Get the number of threads available to the current process.""" | |
| return multiprocessing.cpu_count() | |
| if __name__ == "__main__": | |
| num_cores = get_num_cores() | |
| num_threads = get_num_threads() | |
| print(f"Number of CPU cores: {num_cores}") | |
| print(f"Number of threads available to the current process: {num_threads}") | |
| url = 'https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGML/resolve/main/llama-2-7b-chat.ggmlv3.q2_K.bin' | |
| filename = wget.download(url) | |
| llm2 = Llama(model_path=filename, seed=random.randint(1, 2**31), lora_path="ggml-adapter-model.bin") | |
| filename = wget.download(url) | |
| theme = gr.themes.Soft( | |
| primary_hue=gr.themes.Color("#ededed", "#fee2e2", "#fecaca", "#fca5a5", "#f87171", "#ef4444", "#dc2626", "#b91c1c", "#991b1b", "#7f1d1d", "#6c1e1e"), | |
| neutral_hue="red", | |
| ) | |
| title = """<h1 align="center">Chat with awesome LLAMA 2 CHAT model!</h1><br>""" | |
| with gr.Blocks(theme=theme) as demo: | |
| gr.HTML(title) | |
| gr.HTML("This model is awesome for its size! It is only 20th the size of Chatgpt but is still decent for chatting. However like all models, LLAMA-2-CHAT can hallucinate and provide incorrect information.") | |
| chatbot = gr.Chatbot() | |
| msg = gr.Textbox() | |
| clear = gr.ClearButton([msg, chatbot]) | |
| #instruction = gr.Textbox(label="Instruction", placeholder=) | |
| def user(user_message, history): | |
| return gr.update(value="", interactive=True), history + [[user_message, None]] | |
| def bot(history): | |
| #instruction = history[-1][1] or "" | |
| user_message = history[-1][0] | |
| #token1 = llm.tokenize(b"### Instruction: ") | |
| #token2 = llm.tokenize(instruction.encode()) | |
| #token3 = llm2.tokenize(b"USER: ") | |
| #tokens3 = llm2.tokenize(user_message.encode()) | |
| #token4 = llm2.tokenize(b"\n\n### Response:") | |
| tokens = llm2.tokenize(user_message.encode()) | |
| history[-1][1] = "" | |
| count = 0 | |
| output = "" | |
| for token in llm2.generate(tokens, top_k=50, top_p=0.73, temp=0.72, repeat_penalty=1.1): | |
| text = llm2.detokenize([token]) | |
| output += text.decode() | |
| count += 1 | |
| if count >= 500 or (token == llm2.token_eos()): | |
| break | |
| history[-1][1] += text.decode() | |
| yield history | |
| response = msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then( | |
| bot, chatbot, chatbot | |
| ) | |
| response.then(lambda: gr.update(interactive=True), None, [msg], queue=False) | |
| gr.HTML("Thanks for checking out this app!") | |
| demo.queue() | |
| demo.launch(debug=True) | |