Spaces:
No application file
No application file
Update README.md
Browse files
README.md
CHANGED
|
@@ -5,33 +5,23 @@ colorFrom: blue
|
|
| 5 |
colorTo: indigo
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
| 8 |
-
app_file: app.py
|
| 9 |
---
|
| 10 |
|
| 11 |
-
# Inflexion Lab
|
| 12 |
|
| 13 |
-
**
|
| 14 |
|
| 15 |
-
|
| 16 |
|
| 17 |
-
|
| 18 |
-
To bridge the digital divide for the Kazakh language by developing open-source models, datasets, and tools that enable seamless human-AI interaction. We combine advanced Deep Learning techniques with linguistic precision to create models that truly understand the nuances of the language.
|
| 19 |
-
|
| 20 |
-
### π¬ Key Areas of Research
|
| 21 |
-
* **Automatic Speech Recognition (ASR):** Fine-tuning large-scale models (Whisper) for mixed-language environments (Kazakh/Russian).
|
| 22 |
-
* **Data Engineering:** Syntactic restructuring of raw speech corpora using LLMs (Gemma, Llama) to create high-quality training data.
|
| 23 |
-
* **Large Language Models (LLMs):** Adapting and aligning foundation models for Turkic languages.
|
| 24 |
|
| 25 |
### π₯ Team
|
| 26 |
-
We are a team of engineers and researchers passionate about AI infrastructure and linguistics.
|
| 27 |
|
|
|
|
| 28 |
* **Askhat Sabitkhanov**
|
| 29 |
* **Dias Ilyas**
|
| 30 |
* **Sergey Klimov**
|
| 31 |
|
| 32 |
-
###
|
| 33 |
-
* **Sybyrla (Whisper Large V3)
|
| 34 |
-
* **KSC2 Structured
|
| 35 |
-
|
| 36 |
-
---
|
| 37 |
-
*Open Science. Open Source. Inflexion.*
|
|
|
|
| 5 |
colorTo: indigo
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# Welcome to Inflexion Lab π
|
| 11 |
|
| 12 |
+
This is the organization of **Inflexion Lab**, an AI research and development group dedicated to advancing state-of-the-art **Natural Language Processing (NLP)** and **Automatic Speech Recognition (ASR)** for the Kazakh language.
|
| 13 |
|
| 14 |
+
In this organization, we continuously release fine-tuned foundation models (such as Whisper), syntactically structured open-source datasets, and tools designed to bridge the digital divide for low-resource languages. Our primary focus is solving complex linguistic challenges, including code-switching and punctuation restoration, to build industrial-grade AI infrastructure for Central Asia.
|
| 15 |
|
| 16 |
+
Feel free to explore our **Sybyrla** model or our **Structured KSC2** dataset below!
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
### π₯ Team
|
|
|
|
| 19 |
|
| 20 |
+
We are a specialized team of researchers and engineers:
|
| 21 |
* **Askhat Sabitkhanov**
|
| 22 |
* **Dias Ilyas**
|
| 23 |
* **Sergey Klimov**
|
| 24 |
|
| 25 |
+
### π¬ Featured Work
|
| 26 |
+
* **[Sybyrla (Whisper Large V3)](https://huggingface.co/InflexionLab/sybyrla)**: A robust ASR model achieving ~12% WER, trained on a strategic mix of KSC2 and Russian Common Voice.
|
| 27 |
+
* **[KSC2 Structured](https://huggingface.co/datasets/InflexionLab/ksc2-structured)**: The official ISSAI corpus remastered with Gemma 27B to restore punctuation and capitalization.
|
|
|
|
|
|
|
|
|