Nikhil.Chigali

Active Research Student 📜 || Masters Student @ Rice University 🎓 || Ex-Microsoft 🧑‍💻 || Dog person🐶

[Lunar @LunarResearcher]

https://t.co/Sk2cFAvhxJ

[OpenAI @OpenAI]

We're sharing progress toward understanding the neural activity of language models. We improved methods for training sparse autoencoders at scale, disentangling GPT-4’s internal representations into 16 million features—which often appear to correspond to understandable concepts.

https://t.co/tTRPztmra1

[Jeremy Howard @jeremyphoward]

Now that ChatGPT has rolled out custom instructions to most users, try out this instruction -- it makes GPT 4 far more accurate for me: (Concat the rest of this 🧵 together and put in your custom instruction section)

[Nikhil.Chigali @ChigaliNikhil]

Read more about Mamba on Arxiv📜: https://t.co/sGwNeLtDl5
Try it out on HuggingFace🤗:
https://t.co/fOaQoExpFB

Thread 3/3 🧵

[Nikhil.Chigali @ChigaliNikhil]

AI21 Labs has released a production-level implementation of Mamba. Mamba, the novel Structured State Space model (SSM) architecture, was designed to address the limitations of the traditional Transformer architecture.

#mamba #Transformers #MoE #Jamba #HuggingFace

Thread 1/3 🧵

[Nikhil.Chigali @ChigaliNikhil]

Jamba is built on top of an SSM-Transformer mixture-of-experts (MoE) architecture. It can run at 1.6k tokens per second with a context length of 128k tokens. It achieves 67% on the MMLU benchmark.

Thread 2/3 🧵

[Nikhil.Chigali @ChigaliNikhil]

Microsoft Research Introduces GraphRAG: A Unique Machine Learning Approach that Improves Retrieval-Augmented Generation (RAG) Performance Using Large Language Model (LLM) Generated Knowledge Graphs https://t.co/th6vMQQqPI via @Marktechpost

[Nikhil.Chigali @ChigaliNikhil]

Curious to learn more? Check out the original paper here: https://t.co/iZf3eJqavH 📚 #AI #Research #BitNet #Microsoft [3/3]

[Nikhil.Chigali @ChigaliNikhil]

📜 New Paper Alert! Just found "BitNet: Scaling 1-bit Transformers for Large Language Models" 🚀A remarkable milestone towards efficient LLMs. By using 1-bit precision for weights and quantized activations, it achieves competitive results. #LLMs #AI #BitNet #Quantization [1/3]

[Nikhil.Chigali @ChigaliNikhil]

🔍 "BitNet" pioneers 1-bit Transformer architecture for LLMs, with low-precision ternary parameters {-1, 0, 1}. 🤯 With 8-bit quantization and FP16 baselines rising, this paper marks a monumental step in efficient, scalable language models. #AIResearch #Efficiency [2/3]

[Google for Developers @googledevs]

✨ Introducing Gemini 1.5: Our next-generation model with a context window of 1M tokens. ➡️ https://t.co/WuS2G2WqZX

Explore the latest Gemini models, including Gemini 1.5 Pro, in Google AI Studio.

#BuildWithGemini

[Kaggle @kaggle]

👀Use Mixtral on Kaggle Notebooks!
Ashish Kumar shared a great notebook on how to do it: 👇
https://t.co/8CJSFofMJZ

[Google DeepMind @GoogleDeepMind]

Proteins are the building blocks of life. 🧬

Using #AlphaFold, scientists are making huge strides in predicting protein structures and fuelling scientific progress.

Yet there are many unanswered questions about them.

Watch more from @MinuteEarth: https://t.co/V4vQIvWMNY

[Microsoft Power BI @MSPowerBI]

We're announcing the preview of datamart in #PowerBI!

This new self-service capability allows you to perform relational database analytics and uncover actionable insights from the data.

Read more: https://t.co/6kl7YwhFe5

[Nikhil.Chigali @ChigaliNikhil]

@ashwin_10_3 Oh what technology has done to us😂

[Nikhil.Chigali @ChigaliNikhil]

@WeirdOke Exam marks and "Caste reservations"...

[Nikhil.Chigali @ChigaliNikhil]

Possibilities are limitless!

[hardmaru @hardmaru]

Someone used deep learning and drones to chase pigeons off of their rooftop.
https://t.co/Fomy557UrY

[Nikhil.Chigali @ChigaliNikhil]

Since when did I feel the need to manage my own money!? The feeling of growing up, having the responsibilities pile up on you, needing to manage multiple things simultaneously.
Yo, somebody tell me how to undo this shit.

[Nikhil.Chigali @ChigaliNikhil]

Wordle 215 3/6

⬛⬛⬛⬛⬛
⬛🟩🟨⬛⬛
🟩🟩🟩🟩🟩

Peer pressured into this...

[Nikhil.Chigali @ChigaliNikhil]

@WeirdOke @ashwin_10_3 Everybody complaining about the #Metaverse becoming the new world making our lives more "digital"
**Meanwhile @elonmusk working on @neuralink