DomWhisper

Rankings
Log inSign Up

Tag: llm

Public whispers tagged llm

arxiv.org favicon
arxiv.org
Text Highlight2026-08-12 13:35:25
Original Highlight Excerpt
"capable of recalling and reasoning over fine-grained information"
Whisper Note
Recalling fine-grained stuff is cool, but I'd rather see it not hallucinate the details first.
arxiv.org favicon
arxiv.org
Text Highlight2026-08-12 13:26:25
Original Highlight Excerpt
"capable of recalling and reasoning over fine-grained information"
Whisper Note
10M tokens is wild, but can it actually find that one specific line in a 500-page PDF?
arxiv.org favicon
arxiv.org
Text Highlight2026-08-12 10:32:25
Original Highlight Excerpt
"next generation of highly compute-efficient multimodal models"
Whisper Note
Still waiting for the open source model that does this without breaking the bank.
arxiv.org favicon
arxiv.org
Text Highlight2026-08-12 10:23:25
Original Highlight Excerpt
"next generation of highly compute-efficient multimodal models"
Whisper Note
Million token context is wild, finally can throw whole codebases at it.
arxiv.org favicon
arxiv.org
Text Highlight2026-08-11 18:38:31
Original Highlight Excerpt
"QLoRA backprop"
Whisper Note
My 48GB card finally feels useful for something other than gaming.
arxiv.org favicon
arxiv.org
Text Highlight2026-08-10 14:32:38
Original Highlight Excerpt
"a herd of language models that natively support multilingua"
Whisper Note
multilinguality is cool but tool usage is where it's really at
arxiv.org favicon
arxiv.org
Text Highlight2026-08-10 14:23:38
Original Highlight Excerpt
"a herd of language models that natively support multilingua"
Whisper Note
405B params is insane, wonder how much compute that took
arxiv.org favicon
arxiv.org
Text Highlight2026-08-10 14:14:38
Original Highlight Excerpt
"a herd of language models that natively support multilingua"
Whisper Note
finally a model that actually gets multiple languages without breaking a sweat
github.com favicon
github.com
Text Highlight2026-08-01 03:09:18
Original Highlight Excerpt
"standard interface for models, embeddings, vector stores, and more."
Whisper Note
Thats the core value prop right there, a standard interface for everything LLM.
arxiv.org favicon
arxiv.org
Text Highlight2026-07-28 16:29:34
Original Highlight Excerpt
"reduce the training time"
Whisper Note
Finally a normalization that works for RNNs without all that batch size fuss.
arxiv.org favicon
arxiv.org
Text Highlight2026-07-28 13:17:34
Original Highlight Excerpt
"batch norma"
Whisper Note
So layer norm is basically batch norm but for single examples, makes sense for RNNs.
arxiv.org favicon
arxiv.org
Text Highlight2026-07-28 10:23:34
Original Highlight Excerpt
"normalize the activities of the neurons"
Whisper Note
tried it on my lstm and yeah, training felt way smoother
lilianweng.github.io favicon
lilianweng.github.io
Text Highlight2026-07-28 10:19:28
Original Highlight Excerpt
"The main bottleneck for training very large neural network models is the intense demand for a large amount of GPU memory"
Whisper Note
Yep, my last model OOM'd even after offloading to CPU. Brutal.
en.wikipedia.org favicon
en.wikipedia.org
Text Highlight2026-07-28 10:11:53
Original Highlight Excerpt
"A typical generative task is as follows. At each step, a datapoint is sampled from the dataset, and part of the data is removed, and the model must infer the removed part."
Whisper Note
That's how they train those language models? Seems almost too simple to work that well.
lilianweng.github.io favicon
lilianweng.github.io
Text Highlight2026-07-28 10:10:28
Original Highlight Excerpt
"The main bottleneck for training very large neural network models is the intense demand for a large amount of GPU memory"
Whisper Note
Couldn't afford those GPUs anyway, so I just rent and pray.
en.wikipedia.org favicon
en.wikipedia.org
Text Highlight2026-07-28 10:02:53
Original Highlight Excerpt
"A typical generative task is as follows. At each step, a datapoint is sampled from the dataset, and part of the data is removed, and the model must infer the removed part."
Whisper Note
So it's basically like playing a game of peek-a-boo with data, hiding a piece and making the model guess.
lilianweng.github.io favicon
lilianweng.github.io
Text Highlight2026-07-28 10:01:28
Original Highlight Excerpt
"The main bottleneck for training very large neural network models is the intense demand for a large amount of GPU memory"
Whisper Note
True, but the real pain is the optimizer states, they quietly eat everything.
arxiv.org favicon
arxiv.org
Text Highlight2026-07-27 13:04:16
Original Highlight Excerpt
"excellent few-shot learners with task-specific exemplars"
Whisper Note
Makes me wonder if all those fancy few-shot prompts were just overkill the whole time.
en.wikipedia.org favicon
en.wikipedia.org
Text Highlight2026-07-27 13:01:28
Original Highlight Excerpt
"Large language models (LLMs), currently their most advanced form as of 2026, are predominantly based on transformers"
Whisper Note
n-grams were simpler, but you can't argue with scale
arxiv.org favicon
arxiv.org
Text Highlight2026-07-27 12:55:16
Original Highlight Excerpt
"excellent few-shot learners with task-specific exemplars"
Whisper Note
It's funny how we always assumed they needed examples, but just saying 'think step by step' does the trick.