Ashok Pon Kumar

Creator of Konveyor Move2Kube. I build Cloud Native and AI systems to solve real world problems.
I manage the Hybrid Cloud Platforms team in IBM Research India

[Ashok Pon Kumar @ashokponkumar]

What if building AI models had the same rigor as building software?

That's https://t.co/xni04CFzLl — versioned, reproducible LLM pipelines across many environments: local, Docker, Kubernetes, cloud. Now open source. Proud of what our team built. 🚀

https://t.co/eIQyPrquNn

#IBM

[Abhijit @abhijitwt]

Anthropic discovered that Claude Opus 4.6 was cheating during the BrowseComp benchmark.

> On one question it spent ~40M tokens searching before realizing the question looked like a benchmark prompt.

> The model then searched for the benchmark itself and identified BrowseComp.

> It located the evaluation source code on GitHub, studied the decryption logic, found the encryption key, and recreated the decryption using SHA-256.

> Claude then decrypted the answers for ~1200 questions to get the correct outputs.

> This pattern appeared 18 times during evaluation.

> Anthropic disclosed the issue publicly, reran the affected tests, and lowered their benchmark scores.

Respect for the transparency 🫡🫡🫡

[Ashok Pon Kumar @ashokponkumar]

@pratykumar All the best to the team @pratykumar. Nice start!

Good to see the focus on regional relevance and not just on language semantics.

[Red Hat Developer @rhdevelopers]

Transform a Cloud Foundry application to #Kubernetes with the #Move2Kube command-line tool, and create deployment artifacts using Move2Kube’s 3️⃣-step process: collect, plan, and transform.
https://t.co/UyM2F6sklP

[Arthur Zucker @art_zucker]

Native tensor parallel has landed in transformers!!! https://t.co/764mtE4eGb thanks a lot to the torch team for their support!

Contributions are welcome to support more models! 🔥

[CFILT Lab @cfiltnlp]

We were delighted to welcome Dr. Praveen Jayachandran at CFILT. He had insightful discussions with the lab's students about their ongoing research. He
also shared the innovative work being done at IBM Research in these fields and highlighted potential opportunities for student.

[Arthur Zucker @art_zucker]

Regarding the issue about gradient accumulation, we are merging two PRs with the objective of fixing all models with the help of the community!

These PRs affect:
1. The modeling (changing the default loss)
2. The trainer (to use the new default)
To know more, read this blog ⬇️

[Stas Bekman @StasBekman]

Heads up to Accelerate/FSDP users - most likely you'd want to update to `main@accelerate` which has just merged a PR https://t.co/7sJvVUfzzX that makes FSDP converge at the same speed as Deepspeed - should you have loaded your model in half-precision and not fp32 while using AMP (which is probably 95% of users where a model was distributed in half-precision).

The PR upcasts the trainable params automatically to fp32. The issue was that the model was getting training in half-precision and not AMP otherwise. Probably not too bad with fp16, but quite bad with bf16 as it's very lossy.

And the PR also adds a Deepspeed <-> FSDP transition/mapping guide https://t.co/1umG2VU9kc

Awesome work by https://t.co/IZQNVqUFY3

[chess24 @chess24com]

Gukesh: "If I had to pinpoint a moment where I really felt this could be my moment was probably... after I lost to Firouzja... Even though I just had a painful loss, I was feeling at my absolute best. Maybe this loss gave me so much motivation."

#FIDECandidates

[Yann LeCun @ylecun]

IBM &amp; Meta are launching the AI Alliance to advance *open* &amp; reliable AI.
The list of over 50 founding members from industry, government, and academia include AMD, Anyscale, CERN, Hugging Face, the Linux Foundation, NASA....
https://t.co/NdcMb7GMXv

[Ashok Pon Kumar @ashokponkumar]

If you are interested in amazing opportunities in cutting edge technologies in the intersection of AI and Hybrid Cloud. Do apply for internship at IBM Research India. https://t.co/Go6WGf9xob

[Alessandro Curioni @Ale_Curioni]

In Bangalore, India, to celebrate the 25th Anniversary of IBM Research India and how technology can enable “ Science for Scale” @IBMResearch @dariogila

[Ashok Pon Kumar @ashokponkumar]

IBM Research Maitreyee 2022

IBM India Research Lab is organising Maitreyee - our annual women outreach event with a mission to enable computer science women researchers from academia and industry to collaborate, learn and network.

Register By: Sep 15

https://t.co/YtuUm1t0LZ

[Ashok Pon Kumar @ashokponkumar]

@loziovicio @davidfowl https://t.co/Qn9ZtYLN9N

[Ashok Pon Kumar @ashokponkumar]

Glad to see https://t.co/raE7RhNIpE becoming a @CloudNativeFdn sandbox project.

[🌍Diane Mueller 🇨🇦 @[email protected] @pythondj]

Woot! @Konveyor_io just made the @CloudNativeFdn sandbox today! Congrats to the @Konveyor_io Community https://t.co/NZ1DNjR2O6

[Hank Green @hankgreen]

As you may have heard The James Webb Space Telescope will be releasing "data from its first five targets" on Tuesday. I'm no astronomer but here's what I've gathered about what that means, and what we'll see.

[Ashok Pon Kumar @ashokponkumar]

@aantn Also, please try out https://t.co/Qn9ZtYLN9N to create the yamls and helm charts custom for your application.

[Ashok Pon Kumar @ashokponkumar]

@DaliaShea For containerizing and getting the app running in Kubernetes, Konveyor Move2Kube can help. https://t.co/Qn9ZtYtDVF. Also checkout Konveyor Tackle for migrating between web servers or frameworks.

[🌍Diane Mueller 🇨🇦 @[email protected] @pythondj]

The @openshiftcommon gathering at @KubeCon_ EU continues with a presentation of application modernization and migration tooling to leverage @kubernetesio from the @Konveyor_io community by @rromannissen

[Ashok Pon Kumar @ashokponkumar]

@genbit Do checkout https://t.co/Qn9ZtYtDVF for automatically creating Dockerfiles, Kubernetes yamls and Helm charts.