a barebones Nanogpt, but finetuned on conversational data
all updates in updates.md
Colab link
to view its capabilites, head to colab link, just run the git clone
, pip install
, and prepare.py
, then run chat.py with --init_from=huggingface
Setting context to start the conversation, making a whole new personality for the chatbot
!python chat.py --init_from=huggingface --context=""
This is a fork of Nanogpt, but trained on the data format of a chatbot like chatgpt, with the format inspired by oasst-pythia-12b
<human> ... <endOfText>
<Bot> ... <endOfText>
<human> ... <endOfText>
<Bot> ... <endOfText>
<human> ... <endOfText>
<Bot> ... <endOfText>
garage-bAInd/Open-Platypus
, etcThough probably out of scope for this model, this is something for future models, With many models, you can ask what you were just talking about or to summarize the conversation above. When that is attempted with this model:
Human: Dogecoin is cool
Bot: indeed, very shibe
Human: what were we just talking about?
Bot: me and a friend gave up on mining, but now I can
as we can see, it continues on with a sentence on mining, confirming that it understood the context(GPT2 info) but it does not have the ability to recall. I suspect that has to do with the models data, and that if I were to feed it data like short-context calling and summarization data, it would gain those abilities
Anyone Who can contribute to the repo, please do so, any and all contributions are welcome, simply add a little to the dataset and expand it dataset would be amazing.
I did not make the data dumps/corpuses that make up this data, and can't account for any biases, as the dataset it self is based off the conversations of real people who may or may not have had biases. The model is meant for academic research purposes, and isn't meant for any important or high risk scenarios. Do not follow its advice
for commercial purposes, just take the files input1.txt
through input36.txt
@misc{zheng2023judging,
title={Judging LLM-as-a-judge with MT-Bench and Chatbot Arena},
author={Lianmin Zheng and Wei-Lin Chiang and Ying Sheng and Siyuan Zhuang and Zhanghao Wu and Yonghao Zhuang and Zi Lin and Zhuohan Li and Dacheng Li and Eric. P Xing and Hao Zhang and Joseph E. Gonzalez and Ion Stoica},
year={2023},
eprint={2306.05685},
archivePrefix={arXiv},
primaryClass={cs.CL}
}