omneity

29 ms·

511.
by omneity 3y ago
I see you're opting for AGPL on a codebase that is designed to be embedded as a library. Genuine question, what kind of user did you have in mind when you decided on this license?
512.
by omneity 3y ago
That's true. But assuming the required memory bandwidth is not already maxed out by this, there might still be a narrow but workable "Goldilocks zone" for this technique to be useful.
513.
by omneity 3y ago
Fascinating. The money quote: "OnnxStream can consume even 55x less memory than OnnxRuntime while being only 0.5-2x slower" The trade-off between (V)RAM use and inference time sounds like it could be advantageous in some scenarios
514.
by omneity 3y ago
Two separate terms here, pickling is a serialization method for Python objects (unrelated to AI per se). Read more here: https://docs.python.org/3/library/pickle.html Then "weights" is just referring to
515.
by omneity 3y ago
Optane Resurgence? The tech looks super cool if it does get commercialized.
516.
by omneity 3y ago
Thanks for sharing. That one's cool, but it's not what I'm looking for. This demo here: https://anything.world/showcase/pink-planet-experience I couldn't find a way to access it.
517.
by omneity 3y ago
Does anyone have a direct link to the Pink Sweat$ demo? It doesn't seem to be available from PS' website.
518.
by omneity 3y ago
Can we possibly expect something similar for other neurodegenerative diseases such as Alzheimer's?
519.
by omneity 3y ago
Agreed on overnight batch processing. Although my vision for it is to have a sort of local service that can provide "intelligence" on demand for other apps, which might request it concurrently, at which point double digit throughp
520.
by omneity 3y ago
Which dataset formats are you planning to support? (OpenAI, ShareGPT, Alpaca, Context-based ...)
521.
by omneity 3y ago
Thank you for sharing a well-written and transparent training report! Do you have plans to train/release a fine-tuned 3b chat version or other variants?
522.
by omneity 3y ago
Generating text fast enough for a human to read it imo is only the bare minimum. New classes of use cases would be possible if you could generate 100s of tokens. For example, on-device classification of emails/texts, LLM-powered recomm
523.
by omneity 3y ago
Do you have a link/reference about Mistral's performance? I'm interested to read more about it.
524.
by omneity 3y ago
Does an LLM need to loop back to re-read its input, even in a regular (read non-sliding) context window? Maybe I'm misunderstanding, but doesn't the hidden state solve the "lookup" problem in this case? In the sense that
525.
by omneity 3y ago
Huggingface tutorials are a decent starting point. The HF toolchain is pretty mature and most llm finetuning projects are a wrapper around HF models, HF Trainer and some config templates. An LLM by default would be trained like in the examp
526.
by omneity 3y ago
How do you split your model to train on multiple gpus while pooling VRAM?
527.
by omneity 3y ago
Every now and then I visit the internet without an adblocker to see how bad it’s getting. In this case, there’s barely a paragraph of text between each ad. And even that text is super light on details. It felt like they took the title and m
528.
by omneity 3y ago
This is pretty cool and might replace my use of nomad on my home server.
529.
by omneity 3y ago
Hah, my bad. I guess I confused with something else. Apologies Mr. Miyazaki and wishing you a long and fruitful life! :D
530.
by omneity 3y ago
The beauty of impermanence, masterfully illustrated. Rest in peace mr. Miyazaki and thank you for making the world a better place with your stories.
531.
by omneity 3y ago
This reminds me that car manufacturers in Germany (maybe in other places too) are actually financial services companies with factories attached to them, or so the joke goes. Almost everyone buys new cars through some form of financing, and
532.
by omneity 3y ago
How can you make use of this info? Like which tool allows you to tighten down to a specific torque?
533.
by omneity 3y ago
Not everybody uses these terms, but the closest you’ll find is Romanized Arabic, or Arabizi. And congrats on your progress!
534.
by omneity 3y ago
Other than fine-tuning and RAG, Guidance allows you to constrain the output of an LLM within a grammar, for example to guarantee JSON output 100% of the time. Here's one library to do this https://github.com/guidance-ai
535.
by omneity 3y ago
Axolotl takes a lot of formats, not all of them are in the form of input/output. "Completion" format only takes a single text value per dataset record. Some other formats are in the form of multiple choice answers, etc. Take
536.
by omneity 3y ago
What you described is basically an input/output pair. The input is the sentence so far, and the output is the next token. You build your dataset by splitting the raw text corpus into sentences, paragraphs or documents, and for each of
537.
by omneity 3y ago
Not OP but I work in a similar space. Most likely parsing unstructured data (a superset of NER).
538.
by omneity 3y ago
One of my hopes is that LLM technology can eventually be used to fight technical debt and help us clean up the software stack.
539.
by omneity 3y ago
As the old saying goes, the best way to get help on the internet is to post a wrong answer. In this case, getting a reply straight off the source.
540.
by omneity 3y ago
You can use https://monitoro.co to monitor and extract TOS, and send the changes to a webhook that does whatever you want with the data :) Disclaimer, I work on Monitoro.

More ›