Ramith Hettiarachchi
@ramith.fyi
PhD Student @CMUPittCompBio.bsky.social / @SCSatCMU.bsky.social
Interested in ML for science/Compuational drug discovery/AI-assisted scientific discovery 🤞
from 🇱🇰🫶 https://ramith.fyi
Interested in ML for science/Compuational drug discovery/AI-assisted scientific discovery 🤞
from 🇱🇰🫶 https://ramith.fyi
Installing alphafold3 on the DGX Spark was pretty easy, but I forgot that JAX preallocates gpu memory* (XLA_CLIENT_MEM_FRACTION=0.95)
Forgot to change those settings, and the spark froze without me being able to ssh 😆
*In the spark, it’s unified mem
Gotto test again either reduced reallocation
Forgot to change those settings, and the spark froze without me being able to ssh 😆
*In the spark, it’s unified mem
Gotto test again either reduced reallocation
November 14, 2025 at 12:58 PM
Installing alphafold3 on the DGX Spark was pretty easy, but I forgot that JAX preallocates gpu memory* (XLA_CLIENT_MEM_FRACTION=0.95)
Forgot to change those settings, and the spark froze without me being able to ssh 😆
*In the spark, it’s unified mem
Gotto test again either reduced reallocation
Forgot to change those settings, and the spark froze without me being able to ssh 😆
*In the spark, it’s unified mem
Gotto test again either reduced reallocation
An application asked for the lifetime number of lines coded.. with gemini's help wrote a code to count net lines (.cpp or .py) an author contributed to a repo..
Thinking of polishing it up further:
current version: gist.github.com/ramithuh/aa6...
Thinking of polishing it up further:
current version: gist.github.com/ramithuh/aa6...
November 5, 2025 at 4:49 PM
An application asked for the lifetime number of lines coded.. with gemini's help wrote a code to count net lines (.cpp or .py) an author contributed to a repo..
Thinking of polishing it up further:
current version: gist.github.com/ramithuh/aa6...
Thinking of polishing it up further:
current version: gist.github.com/ramithuh/aa6...
Now supports clusters behind VPNs... (openconnect to the rescue) -- idea is there's a button that lets you copy the vpn connection command (then you use duo etc)..
November 3, 2025 at 4:57 AM
Now supports clusters behind VPNs... (openconnect to the rescue) -- idea is there's a button that lets you copy the vpn connection command (then you use duo etc)..
dgx spark training speed impression while training a diffusion language model on Shakespeare:
DGX Spark: ~1.3 it/s
L40: ~3.5 it/s
(same batch size used here)
DGX Spark: ~1.3 it/s
L40: ~3.5 it/s
(same batch size used here)
November 2, 2025 at 4:38 PM
dgx spark training speed impression while training a diffusion language model on Shakespeare:
DGX Spark: ~1.3 it/s
L40: ~3.5 it/s
(same batch size used here)
DGX Spark: ~1.3 it/s
L40: ~3.5 it/s
(same batch size used here)
this is fun to watch...
October 26, 2025 at 4:35 AM
this is fun to watch...
claude made a nice api spec..
October 26, 2025 at 1:09 AM
claude made a nice api spec..
prototyping a dashboard to submit training jobs to a cluster..
The idea is to do prototyping in a cluster-agnostic manner (local GPU), then once the prototype is ready for a production run, use a dashboard to see which kinds of GPUs are available and submit it..
The idea is to do prototyping in a cluster-agnostic manner (local GPU), then once the prototype is ready for a production run, use a dashboard to see which kinds of GPUs are available and submit it..
October 26, 2025 at 12:58 AM
prototyping a dashboard to submit training jobs to a cluster..
The idea is to do prototyping in a cluster-agnostic manner (local GPU), then once the prototype is ready for a production run, use a dashboard to see which kinds of GPUs are available and submit it..
The idea is to do prototyping in a cluster-agnostic manner (local GPU), then once the prototype is ready for a production run, use a dashboard to see which kinds of GPUs are available and submit it..
when you cant miss out a dinner w friends, yet as a TA you have to answer some questions students have* 🥹😅
*for a HW that's due midnight
*for a HW that's due midnight
October 22, 2025 at 6:26 AM
when you cant miss out a dinner w friends, yet as a TA you have to answer some questions students have* 🥹😅
*for a HW that's due midnight
*for a HW that's due midnight
one of the fondest memories I have during high school time is intel shipping me an edison development board..
back in 2014, i emailed a PM @ Intel with a suggestion, she kindly acknowleged it.. and the next day she asked for my address so that she could send me an intel galileo board..
(1/4)
back in 2014, i emailed a PM @ Intel with a suggestion, she kindly acknowleged it.. and the next day she asked for my address so that she could send me an intel galileo board..
(1/4)
October 21, 2025 at 5:00 AM
one of the fondest memories I have during high school time is intel shipping me an edison development board..
back in 2014, i emailed a PM @ Intel with a suggestion, she kindly acknowleged it.. and the next day she asked for my address so that she could send me an intel galileo board..
(1/4)
back in 2014, i emailed a PM @ Intel with a suggestion, she kindly acknowleged it.. and the next day she asked for my address so that she could send me an intel galileo board..
(1/4)
Sometimes I cook so that I can listen to a podcast
October 19, 2025 at 5:40 PM
Sometimes I cook so that I can listen to a podcast
the package ydata-profiling is pretty nice
github.com/ydataai/ydat...
was doing some data analysis in the plinder dataset, generating a bunch of high level info with a single command
github.com/ydataai/ydat...
was doing some data analysis in the plinder dataset, generating a bunch of high level info with a single command
October 14, 2025 at 3:29 PM
the package ydata-profiling is pretty nice
github.com/ydataai/ydat...
was doing some data analysis in the plinder dataset, generating a bunch of high level info with a single command
github.com/ydataai/ydat...
was doing some data analysis in the plinder dataset, generating a bunch of high level info with a single command
view from Eagle rock trailhead was breathtaking ❤️
September 30, 2025 at 3:47 AM
view from Eagle rock trailhead was breathtaking ❤️
My takeaway:
Their Approach == ESM2 (3B) + SimpleTransformer w. Flow Matching (X B)
When X=100M, it falls way short of ESMFold performance (10-15%)
When X=3B, it closes the gap
---> Still, AF2+MSA (95M) is far superior
But it's nice to know how far simple transformer+flow matching can take us
Their Approach == ESM2 (3B) + SimpleTransformer w. Flow Matching (X B)
When X=100M, it falls way short of ESMFold performance (10-15%)
When X=3B, it closes the gap
---> Still, AF2+MSA (95M) is far superior
But it's nice to know how far simple transformer+flow matching can take us
September 24, 2025 at 4:27 PM
My takeaway:
Their Approach == ESM2 (3B) + SimpleTransformer w. Flow Matching (X B)
When X=100M, it falls way short of ESMFold performance (10-15%)
When X=3B, it closes the gap
---> Still, AF2+MSA (95M) is far superior
But it's nice to know how far simple transformer+flow matching can take us
Their Approach == ESM2 (3B) + SimpleTransformer w. Flow Matching (X B)
When X=100M, it falls way short of ESMFold performance (10-15%)
When X=3B, it closes the gap
---> Still, AF2+MSA (95M) is far superior
But it's nice to know how far simple transformer+flow matching can take us
Good to be back in Pittsburgh ❤️
August 30, 2025 at 12:01 AM
Good to be back in Pittsburgh ❤️
August 18, 2025 at 5:34 PM
haha.. thats pretty cool! 😆
July 13, 2025 at 3:22 PM
haha.. thats pretty cool! 😆
kinda cool that the url has all the code, and nothing is stored on the server..
can even put in a git commit message (to explain something significant underlying the commit) 🤔 (although i'd be wasting 1KB for a commit message)
can even put in a git commit message (to explain something significant underlying the commit) 🤔 (although i'd be wasting 1KB for a commit message)
June 26, 2025 at 5:02 PM
kinda cool that the url has all the code, and nothing is stored on the server..
can even put in a git commit message (to explain something significant underlying the commit) 🤔 (although i'd be wasting 1KB for a commit message)
can even put in a git commit message (to explain something significant underlying the commit) 🤔 (although i'd be wasting 1KB for a commit message)
It's nice to add a keybinding in Jupyter Lab for things like "restart kernel and run all"
June 13, 2025 at 2:10 PM
It's nice to add a keybinding in Jupyter Lab for things like "restart kernel and run all"
damn JVKE has written songs on how everything feels like 😆❤️
May 5, 2025 at 12:00 AM
damn JVKE has written songs on how everything feels like 😆❤️
pymol looks cool with `bg_color [0.0, 0.0, 0.502]`
sort of old computer interface background color scheme
sort of old computer interface background color scheme
April 26, 2025 at 5:12 PM
pymol looks cool with `bg_color [0.0, 0.0, 0.502]`
sort of old computer interface background color scheme
sort of old computer interface background color scheme
Const pH MD going weee 🏃♂️
April 26, 2025 at 4:27 PM
Const pH MD going weee 🏃♂️
ok upgrading deepspeed worked.. special thanks to perplexity and gemini for validating my choices before running them
April 23, 2025 at 4:45 AM
ok upgrading deepspeed worked.. special thanks to perplexity and gemini for validating my choices before running them
Had some fun for homework, running a pipeline composed of virtual screening (pharmit) + pocket conditioned ligand design (DrugFlow --> this was easy to setup/run 💯)
April 12, 2025 at 4:45 AM
Had some fun for homework, running a pipeline composed of virtual screening (pharmit) + pocket conditioned ligand design (DrugFlow --> this was easy to setup/run 💯)