its interesting how MSA is very important for protein structure prediction to extract co-evolution signals, but in the small molecule design/conformer generation scenarios with the scale we are working, the model can "learn physics"
Agree, writing code on own definitely gives the idea of real issues we’d encounter..
and we might not get that picture if we fully relied on coding agents.. like less failures, less learning 😅 it’s hard to balance this
im reorganizing one of my old repo coded in JAX, I'm starting to question whether in the age of claude code, how important is knowing syntax, etc.. 😅
feels like ideas matter more than the implementation at this time, where coding is straightforward if you ask the right fundemental questions
Scared to use less popular vscode extentions etc after reading this
Haha true!
today i learnt that you can filter runs on @weightsbiases.bsky.social using the git commit id
note that the mol* slider is not related to the point i wanted to make.. its just the change of example id runs all cells affected by it
starting to love the @marimo.io notebook's reactive cell execution. all the nodes in the DAG are on the same page!
ok nvm, i'm so stupid.. sometime back I restricted sshd to use only core 0 🤦♂️🤦♂️🤦♂️🤦♂️🤦♂️🤦♂️🤦♂️🤦♂️
bsky.app/profile/rami...
this is kinda stupid.. need to figure out why numpy codes only restrict to one core
python -c "import numpy; import os; print(f'Allowed Cores: {len(os.sched_getaffinity(0))}')"
ran 3 python codes, all 3 were using the same cpu core while 19 were free 🤔 i thought the os is supposed to handle this
First, @tattabio.bsky.social is now on Bluesky!💙 and second, we launched mult-sequence CoSearch on SeqHub!
Agree.. I tested a notebook with some tests at the bottom. Was able to run that file with pytest as well, which is helpful
Also i noticed that the support for pandas dataframes is pretty nice.. (to make quick charts etc) -- not sure if jupyter allowed that
I'm still figuring it out, but in general, I like the idea that pytest can be in the same marimo notebook, so the issues of outdated notebooks might not happen..
another con: I don't like the UI much :/ jupyterhub's tabs feels way better for switching between stuff
Pros:
- Feel like fewer mistakes can happen (Nice that if I change something at the beginning, everything connected gets updated)
- pytest can be in the same .py file
Cons:
- cannot archive things (I used to look at notebooks to understand shapes, visualizations) - marimo does let u export though.
@marimo.io's hover tooltips showing function signatures & docs is nice
I like how @anthropic.com is doing research into the downsides of AI use.. so that we can improve how we interact/get help
www.anthropic.com/research/AI-...
yea! html is still experimental in typst.. noticed a couple of bugs with the math mode when generating the webstite
No prob! I have only used Obsidian for a bit... My major complaint would be 1) lack of proper bibliography formatting options, 2) collaborating on docs?
I think for personal notes obsidian should be good!
I was writing this yesterday, new.ramith.fyi/notes/2026-0..., and was quite impressed
Yeah, atm typst doesn't have a markdown export.. it's designed to replace LaTeX..
need to do more testing to see how well it fits my workflow..
But I like the fact that blog posts can be written with all the powerful Bib + and other formatting tools in typst..
Starting to appreciate the simplicity of this over Ghost CMS that I've been using for quite some time
Screenshot of the Typst web application interface showing a split-screen view of a project titled "typst-website." The left panel displays a file directory with a "posts" folder expanded, revealing various dated entries. The center panel contains a code editor with Typst markup for a document titled "The Normal Distribution," including text, mathematical syntax for the probability density function, and code for generating a diagram. The right panel shows the live preview of the rendered document, featuring formatted text, the mathematical formula clearly typeset, and a blue histogram chart illustrating the distribution. The path bar at the top indicates the file location: Ramith > typst-website > content > posts > 2025-10-30-normal-distribution > index.typ
building a website through @typst.app is kinda straightforward..
was trying out this template: github.com/vsheg/tufted, forum.typst.app/t/tufted-web...
for eg: Below you can see an example post from their template:
new.ramith.fyi/posts/2025-1...
does flash attention variants support arbitrary attention biases 🤔
Taken together, we hope our results highlight the current limitations of the self-consistency evaluation that is so commonly used in the field, and thus encourage to establish new/additional criteria, or at least be more aware of the downsides of AF & co metrics when assessing designs.
18/19
Limitations of the refolding pipeline for de novo protein design
Figure 1
Figure 2
Figure 3
Limitations of the refolding pipeline for de novo protein design [new]
Refolding pipelines eval: evo info can hurt seq-struct compat.
was hoping to see some reasoning traces on the math countdown task.. instead, diffusion LLM, started saying "exactly" too many times 🤨
Maybe its "reasoning" is during denoising (remasking + unveiling), rather than the autoregressive case where it explicitly says ("oops i got that wrong")
This is the question in the validation set
Trying to train a diffusion LLM with reasoning on a math task (without SFT), and this is what it spits out 😒😑
glad this is available on Youtube!
Of course! Feel free to ping me if you have issues setting it up.. idea is to host the dashboard locally, and add cluster configs
I will improve the readme a bit!
