kat33/llama.cpp
2
1import os # to check if file exists2import sys # to flush stdout3import markdown # to render answer4 5import gradio as gr6#import transformers7#from transformers import pipeline8from llama_cpp import Llama9from huggingface_hub import hf_hub_download10 11model_repo="TheBloke/Nous-Hermes-13B-GGML"12model_filename="nous-hermes-13b.ggmlv3.q4_K_S.bin"13 14#model="TheBloke/Nous-Hermes-13B-GGML"15#model="https://huggingface.co/TheBloke/Nous-Hermes-13B-GGML/resolve/main/nous-hermes-13b.ggmlv3.q4_K_S.bin"16 17def download_model():18 # See https://github.com/OpenAccess-AI-Collective/ggml-webui/blob/main/tabbed.py19 file_path="/home/user/.cache/huggingface/hub/models--TheBloke--Nous-Hermes-13B-GGML/snapshots/f1a48f90a07550e1ba30e347b2be69d4fa5e393b/nous-hermes-13b.ggmlv3.q4_K_S.bin"20 if os.path.exists(file_path):21 return file_path22 else:23 print("Downloading model...")24 sys.stdout.flush()25 file = hf_hub_download(26 repo_id=model_repo, filename=model_filename27 )28 print("Downloaded " + file)29 return file30 31def question_answer(context, question, max_tokens):32 mfile=download_model()33 # structure the prompt to make it easier for the ai34 question1="\"\"\"\n" + question + "\n\"\"\"\n"35 text=context + "\n\nQuestion: " + question1 + "\nPlease use markdown formatting for answer. \nAnswer:\n" 36 llm = Llama(model_path=mfile)37 output = llm(text, max_tokens=max_tokens, stop=["### Response"], echo=True)38 print(output)39 40 # remove the context and leave only the answer41 answer=output['choices'][0]['text']42 answer = answer.replace(text, "", 1)43 44 # render the markdown and return the html and question45 html_answer = markdown.markdown(answer)46 return question, html_answer47 '''48 Output is of the form:49 {50 "id": "cmpl-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",51 "object": "text_completion",52 "created": 1679561337,53 "model": "./models/7B/ggml-model.bin",54 "choices": [55 {56 "text": "Q: Name the planets in the solar system? A: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune and Pluto.",57 "index": 0,58 "logprobs": None,59 "finish_reason": "stop"60 }61 ],62 "usage": {63 "prompt_tokens": 14,64 "completion_tokens": 28,65 "total_tokens": 4266 }67 }68 '''69 70 # old transformers code71 #generator = pipeline(model=model, device_map="auto")72 #return generator(text)73 74 75app=gr.Interface(fn=question_answer, inputs=["text", "text",gr.Slider(33, 2333)], outputs=["textbox", "html"])76app.launch()77 