ai4bharat/indic-parler-tts
Text-to-Speech
•
Updated
•
14.6k
•
88
Shoaib Hossain
KillerShoaib
AI & ML interests
Computer Vision: Obj Detection, Segmentation, Generative AI
NLP: LLMs
Recent Activity
liked
a model
about 1 month ago
ai4bharat/indic-parler-tts
liked
a model
2 months ago
turjo4nis/colbertv2.0-bn
Organizations
None yet
KillerShoaib's activity
liked
a
model
about 1 month ago
reacted to
nyuuzyou's
post with 👍
about 2 months ago
Post
2203
🎵 Introducing Suno Music Generation Dataset -
nyuuzyou/suno
Dataset highlights:
- 659,788 AI-generated music samples with comprehensive metadata from suno.com
- Multilingual content with English as primary language, including Japanese and other languages
- Each entry contains rich metadata including:
- Unique song ID, audio/video URLs, and thumbnail images
- AI model version and generation parameters
- Song metadata (tags, prompts, duration)
- Creator information and engagement metrics
- Released to the public domain under Creative Commons Zero (CC0) license
The dataset structure includes detailed information about each generated piece, from technical parameters to user engagement metrics, making it particularly valuable for:
- Music generation model training
- Cross-modal analysis (text-to-audio relationships)
- User engagement studies
- Audio classification tasks
- Music style and genre analysis
Dataset highlights:
- 659,788 AI-generated music samples with comprehensive metadata from suno.com
- Multilingual content with English as primary language, including Japanese and other languages
- Each entry contains rich metadata including:
- Unique song ID, audio/video URLs, and thumbnail images
- AI model version and generation parameters
- Song metadata (tags, prompts, duration)
- Creator information and engagement metrics
- Released to the public domain under Creative Commons Zero (CC0) license
The dataset structure includes detailed information about each generated piece, from technical parameters to user engagement metrics, making it particularly valuable for:
- Music generation model training
- Cross-modal analysis (text-to-audio relationships)
- User engagement studies
- Audio classification tasks
- Music style and genre analysis
liked
a
model
2 months ago
reacted to
MonsterMMORPG's
post with 🔥
3 months ago
Post
3651
Stability AI published their most power newest model Stable Diffusion 3.5 Large. This model unlike FLUX is full model not distilled and has huge potential. I have done extensive research and publishing all of it in this video regarding how to use SD 3.5 Large with the best settings. Moreover, I am sharing how to use FLUX DEV with the best possible configuration as well. Moreover, I am making a huge comparison between SD 3.5 and FLUX and you are going to learn who is the winner.
https://youtu.be/-zOKhoO9a5s
62 Prompts tested on all experiments to find best Sampler + Scheduler for Stable Diffusion 3.5 Large and SD 3.5 Large vs FLUX DEV > https://youtu.be/-zOKhoO9a5s
FLUX Dev vs SD 3.5 Large fully compared.
SD 3.5 Large FP16 vs Scaled FP8 fully compared.
T5 XXL FP8 vs Scaled FP8 vs FP16 fully compared.
FLUX FP16 vs Scaled FP8 fully compared.
Also how to install SwarmUI on Windows, Massed Compute and RunPod shown in the tutorial.
I have shown how to use FLUX and SD 3.5 Large in details as well.
https://youtu.be/-zOKhoO9a5s
62 Prompts tested on all experiments to find best Sampler + Scheduler for Stable Diffusion 3.5 Large and SD 3.5 Large vs FLUX DEV > https://youtu.be/-zOKhoO9a5s
FLUX Dev vs SD 3.5 Large fully compared.
SD 3.5 Large FP16 vs Scaled FP8 fully compared.
T5 XXL FP8 vs Scaled FP8 vs FP16 fully compared.
FLUX FP16 vs Scaled FP8 fully compared.
Also how to install SwarmUI on Windows, Massed Compute and RunPod shown in the tutorial.
I have shown how to use FLUX and SD 3.5 Large in details as well.
liked
2
datasets
3 months ago
reacted to
kz919's
post with 🚀
4 months ago
Post
1329
Just for the meme.
But the clear lesson I learnt from building these demos are, the more powerful the underlying base model is, the closer you will get to GPT4o1. CoT is nothing more than simply inducing the latent reasoning capability from the model.
kz919/GPT4-O1-Proximas
But the clear lesson I learnt from building these demos are, the more powerful the underlying base model is, the closer you will get to GPT4o1. CoT is nothing more than simply inducing the latent reasoning capability from the model.
kz919/GPT4-O1-Proximas
updated
3
models
4 months ago
I need Bangla dataset for pretraining tinyllama .
3
#1 opened 8 months ago
by
ar08
upvoted
an
article
8 months ago
Article
Vision Language Models Explained
•
241
updated
3
models
8 months ago
liked
a
Space
8 months ago
About Dataset
1
#1 opened 8 months ago
by
hasnatz
upvoted
an
article
8 months ago
Article
Fine-Tune Whisper with 🤗 Transformers
•
144
reacted to
Sentdex's
post with 🔥
8 months ago
Post
8484
Okay, first pass over KAN: Kolmogorov–Arnold Networks, it looks very interesting!
Interpretability of KAN model:
May be considered mostly as a safety issue these days, but it can also be used as a form of interaction between the user and a model, as this paper argues and I think they make a valid point here. With MLP, we only interact with the outputs, but KAN is an entirely different paradigm and I find it compelling.
Scalability:
KAN shows better parameter efficiency than MLP. This likely translates also to needing less data. We're already at the point with the frontier LLMs where all the data available from the internet is used + more is made synthetically...so we kind of need something better.
Continual learning:
KAN can handle new input information w/o catastrophic forgetting, which helps to keep a model up to date without relying on some database or retraining.
Sequential data:
This is probably what most people are curious about right now, and KANs are not shown to work with sequential data yet and it's unclear what the best approach might be to make it work well both in training and regarding the interpretability aspect. That said, there's a rich long history of achieving sequential data in variety of ways, so I don't think getting the ball rolling here would be too challenging.
Mostly, I just love a new paradigm and I want to see more!
KAN: Kolmogorov-Arnold Networks (2404.19756)
Interpretability of KAN model:
May be considered mostly as a safety issue these days, but it can also be used as a form of interaction between the user and a model, as this paper argues and I think they make a valid point here. With MLP, we only interact with the outputs, but KAN is an entirely different paradigm and I find it compelling.
Scalability:
KAN shows better parameter efficiency than MLP. This likely translates also to needing less data. We're already at the point with the frontier LLMs where all the data available from the internet is used + more is made synthetically...so we kind of need something better.
Continual learning:
KAN can handle new input information w/o catastrophic forgetting, which helps to keep a model up to date without relying on some database or retraining.
Sequential data:
This is probably what most people are curious about right now, and KANs are not shown to work with sequential data yet and it's unclear what the best approach might be to make it work well both in training and regarding the interpretability aspect. That said, there's a rich long history of achieving sequential data in variety of ways, so I don't think getting the ball rolling here would be too challenging.
Mostly, I just love a new paradigm and I want to see more!
KAN: Kolmogorov-Arnold Networks (2404.19756)
- 5 replies
liked
a
dataset
8 months ago