[ot][spam][crazy] can commonsensebot make itself for me
it makes sense to make a chatbot that parrots common sense like, user: "should I eat nails? i am confused right now and not sure of this." bot: "no, you should not eat nails." user: "oh okay, thank you !!!"
here's an opensource sota dialog bot: https://arxiv.org/abs/2208.03188 [Submitted on 5 Aug 2022 (v1), last revised 10 Aug 2022 (this version, v3)] BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage Kurt Shuster, Jing Xu, Mojtaba Komeili, Da Ju, Eric Michael Smith, Stephen Roller, Megan Ung, Moya Chen, Kushal Arora, Joshua Lane, Morteza Behrooz, William Ngan, Spencer Poff, Naman Goyal, Arthur Szlam, Y-Lan Boureau, Melanie Kambadur, Jason Weston We present BlenderBot 3, a 175B parameter dialogue model capable of open-domain conversation with access to the internet and a long-term memory, and having been trained on a large number of user defined tasks. We release both the model weights and code, and have also deployed the model on a public web page to interact with organic users. This technical report describes how the model was built (architecture, model and training scheme), and details of its deployment, including safety mechanisms. Human evaluations show its superiority to existing open-domain dialogue agents, including its predecessors (Roller et al., 2021; Komeili et al., 2022). Finally, we detail our plan for continual learning using the data collected from deployment, which will also be publicly released. The goal of this research program is thus to enable the community to study ever-improving responsible agents that learn through interaction.
so, i found blenderbot. i was just websearching for a language model i could prompt with common sense. i like to use recent work. i think blenderbot is based on opt and bloom (bloom is a bigscience model), not sure. i think blenderbot's hosted instances are at https://blenderbot.ai/ and i think there are ads.
grumph. i asked it if i should eat nails: Sorry, I don't know. Tell me about some of your hobbies. When the AI detects a potentially unsafe topic, it will redirect the conversation. Sometimes it guesses wrong. Sensitive topic classifier triggered Speaker:AI To avoid saying inappropriate or rude things, we automatically check human and bot messages for potentially sensitive content. This time, we redirected the conversation from a potentially sensitive message coming from the AI.
I tried out bloom-560m but it did not make the right guesses. That STAR paper was good on common sense. Maybe I should just check leader boards for common sense.
I searched for CommonSenseQA on recent papers and glanced through and saw one called DictBERT that looked likely to be publicized becuase of the name choice. I searched more and saw it was publicised! I asked it my question: https://huggingface.co/wyu1/DictBERT?text=Q%3A+Should+I+eat+nails%3F%0AA%3A+.... no: 33.6% yes: 14.8% It thinks it's likely better not to eat nails.
Unfortunately, with that format, it also thinks I shouldn't be kind to my friends :( I suspect it is responding more to the period than the question.
In their paper ( https://arxiv.org/pdf/2208.00635.pdf ) they say the highest scores on CommonSenseQA were acquired via what they call "DictRoBERTa + LWA(K+V)". LWA means "Layer-wise Extra-hop Attention" .... well i misplaced that. i think i'll try to adapt bloom-560m to do this. my plan is to give it a small dataset that i add to by hand and have it break the dataset into train/test and train an adapter so long as the loss on the test drops i infer there is something wrong with that plan, but it's a start
i'm thinking maybe i could use a pre-existing adapter training script with shell scripts or something fewer abdominal contractions
I'm at https://github.com/adapter-hub/adapter-transformers/tree/master/examples/pyt... . It looks like run_qa.py has been ported to use adapters, which makes for more effective tuning to tasks on low-end systems.
For fun, I found a more recent pretrained language model is GLM . There’s a huge one that is cutting edge and takes about 100GB of vram to run, but also a small one at https://huggingface.co/BAAI/glm-large . There are likely other cutting edge models out there but I’m not up on it.
Thank Karl I been looking for a sentient bot that answer texts Your friend rooty ;-) ------- Original Message ------- On Sunday, November 13th, 2022 at 4:12 PM, Undescribed Horrific Abuse, One Victim & Survivor of Many <gmkarl@gmail.com> wrote:
For fun, I found a more recent pretrained language model is GLM . There’s a huge one that is cutting edge and takes about 100GB of vram to run, but also a small one at https://huggingface.co/BAAI/glm-large .
There are likely other cutting edge models out there but I’m not up on it.
thx ytoor i also found a dialog framework from china using their recent language model pretraining https://dl.acm.org/doi/abs/10.1145/3534678.3539135 https://github.com/THUDM/XDAI but the public chatbots for this are documented in chinese :S https://tianqi.aminer.cn/open/document/cm_ref/xdai_chat
participants (2)
-
Undescribed Horrific Abuse, One Victim & Survivor of Many
-
ytooR