Ravi Teja Rachakonda

Video AI at Minivet

[Ravi Teja Rachakonda @tejarr]

Bad grammar makes me [sic].

[Neil @neilshroff]

This was (obviously?) satire, but reality is almost stranger than fiction....read full policy here (https://t.co/k1TDXIG32T).

[Ravi Teja Rachakonda @tejarr]

O brother where wert thou all these days?!

[Charlie Marsh @charliermarsh]

I built a prototype that infers the appropriate CUDA version at runtime and installs the latest compatible PyTorch version (below, 2.6.0+cu126, i.e., CUDA 12.6) without any configuration.

[Ravi Teja Rachakonda @tejarr]

Finally!

[Patrick Loeber @patloeber]

We’ve updated the OpenAI compatibility layer for the Gemini API. You can now use more Gemini models and capabilities in the OpenAI SDK:

- Audio inputs
- Image generation
- Listing models

Here's how you can transcribe files & generate images:

[jason @jxnlco]

Vibe coding is actually slower if you're not stupid

[Aditya Rachakonda @arrac]

If you know any data scientists/ML engineers please forward this message. We have positions up to 40LPA+. Please retweet for visibility.

DM me for the JD.
Please retweet for visibility.

[Minivet AI Labs @MinivetAI]

Kokoro is crazy good and all this is available on a CPU as well.

[Minivet AI Labs @MinivetAI]

China leading OSS AI efforts and blowing away the closed source competition at a fraction of the cost. First Qwen and now Deepseek.

DeepSeek-V3 is a revelation when it comes to cost to train. Interesting times ahead.

[Ravi Teja Rachakonda @tejarr]

Guided generation for LLMs is basically a superpower. Great initiative to create a library of guided output formats: https://t.co/bvMzHogZhl

[Qwen @Alibaba_Qwen]

🎄Happy holidays and we wish you enjoy this year. Before moving to 2025, Qwen has the last gift for you, which is QVQ!

🎉 This may be the first open-weight model for visual reasoning. It is called QVQ, where V stands for vision. It just reads an image and an instruction, starts thinking, reflects while it should, keeps reasoning, and finally it generates its prediction with confidence! However, it is still experimental and this preview version still suffers from a number of limitations (mentioned in our blog), which you should pay attention to while using the model. Feel free to refer to the following links for more information:

* Blog: https://t.co/smYDEeGhCs
* HF: https://t.co/PYfXSFDk3U
* ModelScope: https://t.co/XN0aLrYc16
* Kaggle: https://t.co/PQHVNUwl2b

🚀 It achieves impressive performance in benchmark evaluation, e.g., MMMU, MathVista, etc. But what is more interesting is that it is exciting to see the AI model behaves differently by thinking deeply and reasoning step by step instead of directly providing answers. Yet, it is still a model for preview. It is unstable, it might fall into repetition, it sometimes doesn't follow instruction, etc. We invite you to try the new interesting model and enjoy playing with it! Feel free to shoot us feedback!

[Minivet AI Labs @MinivetAI]

The results look so good and so precise and the claims so out of the world, all while being open source. Great work!

[Ravi Teja Rachakonda @tejarr]

Come to think of it, not just Documents but also Audio and Video too.

[Ravi Teja Rachakonda @tejarr]

I’m sure every document format is going to basically have an “Export to Embeddings” feature soon…

[kepano @kepano]

Microsoft just released a tool that lets you convert Office files to Markdown. Never thought I'd see the day.

Google also added Markdown export to Google Docs a few months ago.

[Ravi Teja Rachakonda @tejarr]

@MinivetAI Trading blows with Qwen in all benchmarks (at all parameter levels) is very impressive, especially when most labs don't even compare themselves against Qwen.

[Ravi Teja Rachakonda @tejarr]

@arrac "Real AI is when machines can ___ like humans."
1. play chess/go ✅
2. write/speak/translate✅
3. draw✅
4. code ✅
5. reason <-- we are here
6. move/act
7. fight??

[Ravi Teja Rachakonda @tejarr]

@MinivetAI Llama 3.3 scoring the highest on IFEval (steerability) is what makes me super excited to try it out.

[Aditya Rachakonda @arrac]

XGrammar looks quite promising. Wish this gets @vllm_project integration soon.

Would be interesting to compare it with other backends like outlines and lm-format-enforcer.

[Ravi Teja Rachakonda @tejarr]

Well, there's always a first time for everything... https://t.co/rzi7h0zVa7

[Ravi Teja Rachakonda @tejarr]

We judge ourselves by our internal motives, and others by their outward actions. http://t.co/zukLlMoLOl

[Ravi Teja Rachakonda @tejarr]

Ivan Pavlov walks into a bar. The bartender rings the bell for last drinks, and he rushes out saying "Damn - I forgot to feed the dog".

[Ravi Teja Rachakonda @tejarr]

Someone on reddit described their idea of Hell in 3 words as thus:
"Immortality, Immobility, Insomnia".

Ouch.