HackerTrans
TopNewTrendsCommentsPastAskShowJobs

michaelnny

no profile record

Submissions

Show HN: Fine-tune llama3 model to support function calling

github.com
3 points·by michaelnny·2 tahun yang lalu·5 comments

Ask HN: The correct prompt/response format to fine-tune LLM for function call

1 points·by michaelnny·2 tahun yang lalu·1 comments

America Is Getting Ready for Space Warfare

wsj.com
3 points·by michaelnny·2 tahun yang lalu·1 comments

Ask HN: How do you handle routing/authorization for large-scale React project?

1 points·by michaelnny·2 tahun yang lalu·2 comments

Show HN: QLoRA-LLM – A custom implementation of QLoRA for fine-tuning a LLM

vectortheta.com
1 points·by michaelnny·3 tahun yang lalu·0 comments

comments

michaelnny
·2 tahun yang lalu·discuss
An honest question, what’s the news on Wordpress episode? Didn’t follow the topic
michaelnny
·2 tahun yang lalu·discuss
One important aspect of the success of AlphaGo and its successor is the game environment is closed domain, and has a stable reward function. With this we can guide the agent to do MCTS search and planning for the best move in every state.

However, such reward system is not available for LLM in an open domain setting.
michaelnny
·2 tahun yang lalu·discuss
Wow, this is a great idea!!! Seriously!
michaelnny
·2 tahun yang lalu·discuss
The company I work for try to add AI/LLM on everything, instead of trying to improve/fix the underlying problem, they now just add the magic AI and everything is “perfect” now.

As an ML engineer and AI developer, I don’t see the real value at all, not to mention the added cost of using LLM
michaelnny
·2 tahun yang lalu·discuss
Maybe they have no idea what’s the difference between “scraping” and “mining”? I mean for non-tech people, these are just buzzwords…
michaelnny
·2 tahun yang lalu·discuss
I looked at the supposed “research” article, there’s nothing to read except few charts to show off the “improvements” over current models. No discussion of the training method or dataset whatsoever.

If I remember correctly, the last decent research paper the company published was probably the InstructGPT paper.
michaelnny
·2 tahun yang lalu·discuss
I’m curious where did you get this information?
michaelnny
·2 tahun yang lalu·discuss
I thought this was a joke at first, but then realized it totally made sense ;)
michaelnny
·2 tahun yang lalu·discuss
This there any sources on the closed model you just mentioned? Also, what do you mean “into a Python equivalent?
michaelnny
·2 tahun yang lalu·discuss
Anyone know how to use GraphRAG to build the knowledge graph on a large collection of private documents, where some might have complex structure (tables, links to other docs), and the content or terms in one document could be related to other documents as well?
michaelnny
·2 tahun yang lalu·discuss
I’ve deleted my reddit account earlier this year and I have no regret. Reddit has turned into something that’s unusable for me.
michaelnny
·2 tahun yang lalu·discuss
This is an interesting and well-timed, as I’m in the need to learn more about centralized logging and monitoring for my job.

I’m also wondering how does Uber handle structured logging
michaelnny
·2 tahun yang lalu·discuss
In case anyone is interested, we have a very simple notebook which try to trick GPT-4o to reveal the structure of the prompt/response related to function calling, and it's located at:

https://github.com/michaelnny/Llama3-FunctionCalling/blob/ma...
michaelnny
·2 tahun yang lalu·discuss
I'm wondering if the tensor parallel settings have any impact on the performance. My naive guess is yes but not sure.

According to the article: """ AMD Configuration: Tensor parallelism set to 1 (tp=1), since we can fit the entire model Mixtral 8x7B in a single MI300X’s 192GB of VRAM.

NVIDIA Configuration: Tensor parallelism set to 2 (tp=2), which is required to fit Mixtral 8x7B in two H100’s 80GB VRAM. """
michaelnny
·2 tahun yang lalu·discuss
Sorry forgot to change repo visibility, it's now public
michaelnny
·2 tahun yang lalu·discuss
Sorry forgot to change repo visibility, it's now public
michaelnny
·2 tahun yang lalu·discuss
Hi, just want to state that we would really appreciate if you can give us some advice or suggestions on this question!
michaelnny
·2 tahun yang lalu·discuss
Just to clarify, I mean I won't try to use vim to write the code.

I definitely recommend people to try the CS50 course.
michaelnny
·2 tahun yang lalu·discuss
I enjoyed the CS50 course few years ago, where they talked about how to build native apps with React Native.

I remember the professor/lecturer was using vim to write the code, no IDE whatsoever! That’s was very cool but I don’t think I’ll ever try it.
michaelnny
·2 tahun yang lalu·discuss
I think the main problem is the way we turn the raw mathematics symbols or equations into tokens, and these suboptimal tokenization may decreases the performance