Link to the paper: https://
github.com/deepseek-ai/De
epSeek-OCR/blob/main/DeepSeek_OCR_paper.pdf
… My "Understanding Multimodal LLMs" article with more info on how images are fed to LLMs, how cross-attention works, etc: https://
magazine.sebastianraschka.com/p/understandin
g-multimodal-llms?utm_source=publication-search
…
@rasbt
-
DeepSeek OCR Paper and Multimodal LLM Article Resources
By
–
-
DeepSeek-OCR Release: Vision Model Capabilities Explained
By
–
DeepSeek finally released a new model and paper. And because this DeepSeek-OCR release is a bit different from what everyone expected, and DeepSeek releases are generally a big deal, I wanted to do a brief explainer of what it is all about. In short, they explore how vision
-
Visual Tokenization Challenges: Aspect Ratios and Image Preprocessing Complexity
By
–
I know it’s popular to hate tokenizers, but visual representations (which are also tokenized) bring a lot of messiness as well. Aspect ratios, cropping, resolution, brightness, etc. Sure, models learn to deal with that but it requires lots of data to make them robust wrt these.
-
Meet in the Middle: Running Two LLMs with Parameter Sharing
By
–
This made me think of the "Meet in the Middle" paper https://
x.com/rasbt/status/1
638538494887821313
…
When I remember correctly, they run two LLMs in both directions with parameter sharing. So it shouldn't impact training time. Kind of wild but hey why not. -
Reading Papers Thoroughly and Training Base Models
By
–
Thanks for sharing! I need to read the paper in more detail so I don't ask these questions that are maybe already answered in the paper . Regarding your second point, that's a good idea. I think this is especially relevant because many based models (like Qwen3) are trained on
-
Dishonesty in Benchmark Reporting: A Corporate Risk
By
–
It's a useful skill but it's still a red flag if you are dishonest about it. Imagine asking the person to report benchmark performance of the LLM that is being developed, would you trust the results? This can backfire on the whole company.
-
Different Model Requires Separate Book Instead of Addition
By
–
Thanks for suggesting. But since this is a very different model, this would not be possible. Adding 5 chapters on implementing this, training this, evaluating this, etc would be a whole other book.
-
Challenges in Model Evaluation and Elo Rating Leaderboards
By
–
Good question. I think the challenge here is that the evaluation is more challenging because of a) it's relative to all the other models
b) it's expensive to run the other models PS: Fun fact, I just wrote a tutorial on Elo ratings and constructing a leaderboard last week: -
TRM motivation and specialized model design considerations
By
–
Thanks for sharing. Points a 1 and 2 are actually what motivated the work on TRM as mentioned in the paper. I think point 3 makes sense as this is a small, specialized model, not a large generalist model.
Point number 4 also makes sense, this has always been true for ML and DL. -
Neural Network Implementation in NumPy with Manual Backpropagation
By
–
Been there, done that . But how about a a neural net in NumPy with manual backprop? https://
github.com/rasbt/python-m
achine-learning-book/blob/master/code/ch12/neuralnet.py
… That was 10 years ago, makes me feel so old now