Spaces:
Configuration error
Configuration error
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,78 @@
|
|
| 1 |
---
|
| 2 |
-
|
|
|
|
| 3 |
emoji: 🐠
|
| 4 |
colorFrom: pink
|
| 5 |
colorTo: yellow
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
-
|
|
|
|
| 1 |
---
|
| 2 |
+
|
| 3 |
+
title: BluDot Labs
|
| 4 |
emoji: 🐠
|
| 5 |
colorFrom: pink
|
| 6 |
colorTo: yellow
|
| 7 |
sdk: static
|
| 8 |
pinned: false
|
| 9 |
+
-------------
|
| 10 |
+
|
| 11 |
+
# BluDot Labs 🐠
|
| 12 |
+
|
| 13 |
+
**Independent AI research lab based in Chennai, India 🇮🇳**
|
| 14 |
+
|
| 15 |
+
[BluDot Labs](https://bludotlabs.com) is a small independent lab working on efficient AI systems, language models, reasoning methods, agent systems, and experimental computing hardware.
|
| 16 |
+
|
| 17 |
+
Most of our work is focused on one question:
|
| 18 |
+
|
| 19 |
+
> How much capability can we get from small models and limited compute?
|
| 20 |
+
|
| 21 |
+
## Research
|
| 22 |
+
|
| 23 |
+
We currently work on:
|
| 24 |
+
|
| 25 |
+
* Small and efficient language models
|
| 26 |
+
* Knowledge distillation
|
| 27 |
+
* Reasoning and reinforcement learning
|
| 28 |
+
* Coding and debugging models
|
| 29 |
+
* Model compression
|
| 30 |
+
* Multi-agent systems
|
| 31 |
+
* Optimization methods
|
| 32 |
+
* AI accelerators and experimental hardware
|
| 33 |
+
* RISC-V and custom compute architectures
|
| 34 |
+
|
| 35 |
+
## Models
|
| 36 |
+
|
| 37 |
+
### Kaveri
|
| 38 |
+
|
| 39 |
+
Kaveri is a family of compact language models focused on coding, reasoning, and post-training research.
|
| 40 |
+
|
| 41 |
+
The project explores different ways of improving smaller models through distillation, reinforcement learning, reward design, and specialized training.
|
| 42 |
+
|
| 43 |
+
### Brahmaputra
|
| 44 |
+
|
| 45 |
+
Brahmaputra is an experimental model family focused on knowledge distillation and capability transfer.
|
| 46 |
+
|
| 47 |
+
The goal is to study how much useful knowledge and reasoning can be transferred from larger models into much smaller ones.
|
| 48 |
+
|
| 49 |
+
## Hardware and Systems
|
| 50 |
+
|
| 51 |
+
BluDot Labs also works on projects outside language models.
|
| 52 |
+
|
| 53 |
+
This includes experimental GPU designs, AI accelerators, distributed compute systems, custom instruction set ideas, and open hardware prototypes.
|
| 54 |
+
|
| 55 |
+
We are interested in the full stack, from training algorithms and models to the hardware that runs them.
|
| 56 |
+
|
| 57 |
+
## Open Research
|
| 58 |
+
|
| 59 |
+
A lot of our work is experimental.
|
| 60 |
+
|
| 61 |
+
Some experiments improve the model. Some do not. We try to document both where possible.
|
| 62 |
+
|
| 63 |
+
Models, benchmarks, code, and research experiments are published so that others can test them, reproduce them, or build on top of them.
|
| 64 |
+
|
| 65 |
+
## Chennai, India
|
| 66 |
+
|
| 67 |
+
BluDot Labs is based in Chennai, Tamil Nadu.
|
| 68 |
+
|
| 69 |
+
We want to contribute to open AI and computing research coming out of India, especially work that can be done by small teams without access to massive amounts of compute.
|
| 70 |
+
|
| 71 |
+
## BluDot Labs
|
| 72 |
+
|
| 73 |
+
**Small lab. Serious experiments.**
|
| 74 |
+
|
| 75 |
+
Website: https://bludotlabs.com
|
| 76 |
+
Location: Chennai, India
|
| 77 |
|
| 78 |
+
Individual repositories may have different licenses, limitations, benchmark setups, and intended uses. Please check the respective model or project page before using them.
|