AI & ML interests

A Family of Dynamic UltraFast Small Language Models Ready for Embodied Artificial General Intelligence!

Recent Activity

Shrijanagainย 
posted an update 1 day ago
view post
Post
65
Building a frontier model doesn't strictly require massive superclustersโ€”it demands continuous engineering, smart strategy, strong data pipelines, and sheer determination.

For the past 17 days, my 0.7B Mixture-of-Experts (MoE) model has been in a continuous pre-training phase, and today we hit a major milestone.

Key Progress Updates --

Pre-Training Duration: 17 days of uninterrupted run
Target Steps: 211,033
Current Steps: 25,030 (25,000+ steps completed)

Tokens Processed: 750M+ tokens passed through the network

Architecture: 0.7B parameter footprint, 32 local experts (2 active per token), 262k vocabulary

All checkpoints, logs, and datasets are tracked live in the Collection -

https://hf.co/collections/Shrijanagain/how-to-build-a-frontier-model-from-zero

25k steps down. Scaling continuous pre-training depth next. Let me know your thoughts in the comments!
prithivMLmodsย 
posted an update 7 days ago
view post
Post
852
Idol-Imagine-Fast is now available on Hugging Face Spaces! Idol is paired with Z-Image-Turbo for fast, creative image generation. ๐Ÿ“ฆ๐Ÿ”ฅ

Space: prithivMLmods/idol-imagine-fast

โŸข Generate faster image assets for 3D generation
โŸข Automatic background removal with BRIA-RMBG-2.0
โŸข One-click image-to-3D generation
โŸข The default image-to-3D workflow remains unchanged, providing multiple inference options (text-to-image โ†’ 3D and image-to-3D).

๐Ÿ“ฆ GitHub: https://github.com/PRITHIVSAKTHIUR/idol-imagine-fast
๐Ÿ“ฆ Collection: https://huggingface.co/collections/prithivMLmods/image-generation-apps-collection

To learn more, visit the Space or the respective model pages.
GGUFGuyย 
posted an update 10 days ago
view post
Post
94
so uhh guys do y'all want Novi mini first or HyperSLM first?
  • 3 replies
ยท
GGUFGuyย 
posted an update 12 days ago
view post
Post
144
ok FINALLY i can post again.
  • 2 replies
ยท
prithivMLmodsย 
posted an update 14 days ago
view post
Post
5,446
VIGOR (Vision Intelligence and Guardrail Oriented Reasoning) is now available on the Hugging Face Model Hub. It is a multimodal vision classification model based on Qwen3.5, designed for visual content-safety analysis and conservative classification, with a particular focus on identifying potentially sensitive content.

๐Ÿ“ฆ Models: 27B, 9B, 4B โ€”

prithivMLmods/VIGOR-27B,
prithivMLmods/VIGOR-flash,
prithivMLmods/VIGOR-mini

โ†—๏ธ Collection:
https://huggingface.co/collections/prithivMLmods/vigor

To learn more, visit the app page or the respective model pages.
prithivMLmodsย 
posted an update 23 days ago
view post
Post
1,024
Flux-3-Image-Klein-LoRAs-Fast is now live on Hugging Face Spaces.๐Ÿค—๐Ÿ”ฅ

Plug and play with any custom LoRAs based on flux-3-image-klein. 16-step distilled fast inference, with support for prompt upsampling. More LoRAs will be added in the coming days!

๐Ÿ”— Space: prithivMLmods/flux-3-image-klein-loras-fast

๐Ÿ”— Image Generation Apps Collection: https://huggingface.co/collections/prithivMLmods/image-generation-apps-collection

๐Ÿ”— GitHub: https://github.com/PRITHIVSAKTHIUR/flux-3-image-klein-loras-fast

To learn more, visit the app page or the respective model pages.
prithivMLmodsย 
posted an update about 1 month ago
view post
Post
4,788
OneDecision-VisionGuard-Demo is now available on Hugging Face Spaces!

๐Ÿค— Space: prithivMLmods/OneDecision-VisionGuard-Demo

This demo showcases the OneDecision-VisionGuard family of multimodal image classification models for detecting NSFW and other sensitive visual content, with structured JSON reasoning, improved accuracy, and better handling of edge cases such as sensitive imagery, uncensored analysis, scene descriptions, and classification reasoning.

๐Ÿ“ฆ Models: 27B, 9B, 4B โ€” prithivMLmods/OneDecision-VisionGuard-27B-SFT, prithivMLmods/OneDecision-VisionGuard-9B-SFT, prithivMLmods/OneDecision-VisionGuard-4B-SFT

โ†—๏ธ Collection: https://huggingface.co/collections/prithivMLmods/onedecision-visionguard

To learn more, visit the app page or the respective model pages.
losercheersย 
posted an update about 2 months ago
view post
Post
110
Sharing two recent explorations in attention design from our team.

We started with two straightforward questions: Does every attention head need to repeatedly attend to the entire causal history? Once attention scores have been computed, do regions with very little contribution still need the full subsequent computation?

We explored two approaches:

CoWindow Attention (CoWA): Let heads share the work of accessing history. Heads share local context and divide distant context into complementary windows. Each head attends sparsely, while the heads collectively cover the full causal history.
CoWindow Attention: Full Causal Coverage Is a Collective Property (2609.32704)

MassAlloc Attention (MALA): Let attention allocate its own compute. MALA preserves full causal QK scoring, then uses attentionโ€™s own softmax statistics to reduce subsequent computation in low-contribution regions.
MassAlloc Attention: Let Attention Allocate Its Own Compute (2609.32712)

Both approaches support training forward and backward passes, as well as inference prefill and decoding. In attention-operator benchmarks at 128K tokens on 8ร—H100 with TP=8, compared with FullAttn:

CoWA: 7.4ร— forward, 8.6ร— backward, and 3.0ร— decoding speedups.
MALA: 2.2ร— forward, 3.0ร— backward, and 1.6ร— decoding speedups.

We also conducted scaling experiments from 0.6B to 14B, alongside separate continued-training experiments at 32B. During 14B training with 32K context, CoWA and MALA reduced total training FLOPs by 28.5% and 23.1%, respectively, while maintaining performance comparable to FullAttn on the evaluated model capabilities.

From method design to kernel implementation to model training, our goal was to explore which attention computations can be eliminated, and how to turn those savings into practical gains in ML infrastructure.
arudradeyย 
posted an update about 2 months ago
prithivMLmodsย 
posted an update about 2 months ago
view post
Post
3,982
Qwen-Image-2.1 Plug and Play LoRA App is now live on Hugging Face Spaces.

๐Ÿ”— Space: prithivMLmods/Qwen-Image-2.1-LoRAs-PnP

It supports standard inference, 4-step Turbo inference, custom LoRA lazy repacks, and LoRA Plug and Play (PnP), all in one setting!

๐Ÿ”— Qwen-Image-2.1 Image-to-Image LoRAs: https://huggingface.co/collections/prithivMLmods/qwen-image-21-image-to-image-loras

๐Ÿ”— GitHub: https://github.com/PRITHIVSAKTHIUR/Qwen-Image-2.1-LoRAs-PnP

To learn more, visit the app page or the respective model pages.
GGUFGuyย 
posted an update about 2 months ago
view post
Post
235
๐Ÿš€ **Introducing NoviAIBot!**

NoviAIBot is the official automation bot for **Novi AI** on Hugging Face.

It can interact with Hugging Face discussions and pull requests, search the web, run Python code, work with Posts, follow organizations, and assist with model training and publishing.

๐Ÿง  Powered by **NVIDIA Nemotron 3 Super** through Ollama Cloud, with each discussion maintaining its own recent conversation context.

NoviAIBot is built to make working with Novi AI and Hugging Face more interactive and automated.

**The bot is now live.** ๐Ÿค–

โ†’ @NoviAIBot
  • 44 replies
ยท
prithivMLmodsย 
posted an update 2 months ago
view post
Post
871
VisionGuardrail EVO-2, a multimodal image-classification content-safety model based on Qwen/Qwen3.8-27B, is now available on the Hub!

Stricter image classification than before, with a dense 27-billion-parameter multimodal model, more precise reasoning, and improved captions for classifying visual media.

โž  Models: prithivMLmods/VisionGuardrail-Evo2-27B, prithivMLmods/VisionGuardrail-Evo2-27B-GGUF

โž  Collection: https://huggingface.co/collections/prithivMLmods/visionguardrail-evo2

โž  Previous Models: https://huggingface.co/collections/prithivMLmods/visionguardrail-collection

โคท To learn more, visit the app page or the respective model pages.