csabakecskemeti (Csaba Kecskemeti)

reacted to MoritzLaurer's post with 👍 4 days ago

Post

2248

Quite excited by the ModernBERT release! 0.15/0.4B small, 2T modern pre-training data and tokenizer with code, 8k context window, great efficient model for embeddings & classification!

This will probably be the basis for many future SOTA encoders! And I can finally stop using DeBERTav3 from 2021 :D

Congrats @answerdotai , @LightOnIO and collaborators like @tomaarsen !

Paper and models here 👇https://huggingface.co/collections/answerdotai/modernbert-67627ad707a4acbf33c41deb

replied to luigi12345's post 5 days ago

I found them here:
https://huggingface.co/GoodiesHere

posted an update 7 days ago

Post

1187

tiiuae Falcon3 10B Q8 playground:
DevQuasar/Mi50

Also find my tiiuae Falcon3 Quant collection here:
https://huggingface.co/collections/DevQuasar/tiiuae-falcon3-676236626f3c57d1a19c6c1d

Enjoy!

reacted to cutechicken's post with ❤️ 7 days ago

Post

2820

🚀 RAGOndevice: High-Performance Local AI Document Analysis Assistant
💫 Core Value
RAGOndevice is a high-performance AI system running locally without cloud dependency. Using CohereForAI's optimized 7B model, it enables professional-grade document analysis on standard PCs. ✨
🌟 Ondevice AI Advantages
1. 🔋 Efficient Resource Utilization

🎯 Optimized 7B Model: Runs on standard PCs
⚡ Local Processing: Instant response without cloud
💻 Low-Spec Compatible: Performs well on regular GPUs
🔄 Optimized Memory: Ensures stable operation

2. 🛡️ Data Security & Cost Efficiency

🔒 Complete Privacy: No external data transmission
🌐 Offline Operation: No internet required
💰 No Subscription: One-time installation
⚙️ Resource Optimization: Uses existing hardware

🎮 Key Features
1. 📊 Powerful Document Analysis

📁 Multi-Format Support: TXT, CSV, PDF, Parquet
🧠 Intelligent Analysis: Automatic structure recognition
👁️ OCR Support: Advanced PDF text extraction
💬 Real-time Chat: Natural language interaction

2. 🔍 Local RAG System

🎯 Efficient Search: TF-IDF based local search
🧩 Context Understanding: Accurate information retrieval
📚 Wikipedia Integration: Rich background knowledge

🎯 Use Cases

🏢 Enterprise: Secure confidential document processing
🔬 Personal Research: Private data analysis
📚 Education: Personal learning material analysis
💻 Development: Local codebase analysis

⭐ Differentiators

🏃‍♂️ Independent Operation: Zero cloud dependency
⚡ Instant Response: No network latency
🔐 Complete Security: Full data control
💎 Cost Efficiency: No ongoing costs

🔮 Future Plans

🚀 Enhanced model optimization
📚 Local knowledge base expansion
⚡ Hardware optimization
📁 Extended file support

🌟 RAGOndevice democratizes high-performance AI, providing the optimal local AI solution for security-sensitive environments. 🚀

🔥 Power of Local AI: Experience enterprise-grade AI capabilities right on your device!

VIDraft/RAGOndevice

reacted to nicolay-r's post with 👀 10 days ago

Post

1921

📢For those who wish to quick start with reasoning / cot application over rows of tabular data but with minimal dependencies, this post would be valuable.

🔎 I found that the problem is that given a bulk of Chain-of-Though (CoT) 🔗 queries for remotely accessed LLM 🤖 (like openrouter / Replicate / OpenAI) might result in connection loss which may lead exception 💥 and challenges with generated content restoration.

Here, is where I contribute with the bulk-chain.
⭐ https://github.com/nicolay-r/bulk-chain

Currently working on 0.24.3 version, in which I am happy to announce the API for developing your apps that are based on CoT schema declaration in JSON (details in attached images 📸)

All you have to do is:
✅ 1. Declare CoT-schema in json
✅ 2. Declare the model or use the preset
✅ 3. Launch code

One example is to use ReplicateIO provider:
https://github.com/nicolay-r/bulk-chain/blob/master/ext/replicate.py

Each model has a wrapped call for inference in try-catch block

posted an update 10 days ago

Post

4460

The AMD Instinct MI50 (~$110) is surprisingly fast for inference Quantized models.

This runs a Llama 3.1 8B Q8 with Llama.cpp
DevQuasar/Mi50

A little blogpost about the HW
http://devquasar.com/uncategorized/amd-radeon-instinct-mi50-cheap-inference/

reacted to cutechicken's post with 🚀 10 days ago

Post

3460

🎮 Introduction to the World's First 3D Tank Game Created Solely with Generative AI 🚀
The advancement of AI technology is revolutionizing game development paradigms. I embarked on a challenge to create a 3D tank game using "only AI assistance," pushing the boundaries of what's possible in AI-driven game development. 🤖
Following the success of my first 2D tank game ( cutechicken/tankwar) 🎯, I ventured into the more challenging realm of 3D FPS game development. Remarkably, using Hugging Face's AI tool ( VIDraft/mouse1), the basic game framework was generated in just one minute ⚡. The 3D modeling ( ginipick/SORA-3D) and sound effects ( fantaxy/Sound-AI-SFX) were also easily created with AI assistance.
The resulting game ( cutechicken/TankWar3D) represents arguably the world's first 3D FPS game created primarily with generative AI. 90% was accomplished through AI capabilities, with the remaining 10% comprising my post-processing work. 🎉
Key Technical Features: 🛠️

Complete 3D rendering system using Three.js 🖥️
Real-time physics-based collision detection and handling 💥
Dynamic shadow and lighting system ☀️
Real-time radar and enemy tracking system 🎯
Advanced particle effects system (explosions, smoke, fire) 💫
Dynamic sound system (engine, firing, explosion sounds) 🔊
AI-driven enemy strategy system (pursuit, evasion, combat) 🤖
Terrain-based tank tilt adjustment 🌍
Real-time crosshair targeting system 🎯
Dynamic UI system (health bars, ammo, score) 📊

Technical Implementation: ⚙️

Physics Engine: 🎳
Custom collision detection system
Dynamic obstacle handling
Real-time terrain interaction

AI Systems: 🧠
State-based AI behavior patterns
Dynamic pathfinding
Tactical decision-making system

Graphics: 🎨
PBR-based rendering
Dynamic particle system
Real-time shadow mapping

reacted to AdinaY's post with 🔥 19 days ago

Post

1571

Sailor 2 🚢 open multilingual model for Southeast Asia by Sea AI Lab🔥
https://huggingface.co/sailor2
sail/Sailor2-20B-Chat

✨ Fully open code & ALL datasets 🙌
✨ 1B/ 8B/20B base & chat expanded on Qwen2.5
✨ Apache 2.0
✨ Supports 15 languages including English, Chinese, Burmese, Cebuano, Ilocano, Indonesian, Javanese, Khmer, Lao, Malay, Sundanese, Tagalog, Thai, Vietnamese, and Waray🇬🇧🇨🇳🇱🇦🇲🇾🇲🇲🇻🇳🇹🇭

replied to their post 24 days ago

self eval results posted soon, partial:

Tasks	Version.	Filter.	n-shot	Metric		Value		Stderr
hellaswag	1	none	0	acc	↑	0.5141	±	0.0050
		none	0	acc_norm	↑	0.6792	±	0.0047

reacted to julien-c's post with 👍 24 days ago

Post

2179

wow 😮

INTELLECT-1 is the first collaboratively trained 10 billion parameter language model trained from scratch on 1 trillion tokens of English text and code.

PrimeIntellect/INTELLECT-1-Instruct

posted an update 25 days ago

Post

1160

Fine Tuned a Llama3.2 3B on the MS Orca-Agents dataset for Analytical-Reasoning
r=16, Alpha=32

If you want to give it a try:

Model:
DevQuasar/analytical_reasoning_r16a32_unsloth-Llama-3.2-3B-Instruct-bnb-4bit

Adapter:
DevQuasar/analytical_reasoning_r16a32_unsloth-Llama-3.2-3B-Instruct-bnb-4bit_adapter

Quants:
DevQuasar/analytical_reasoning_r16a32_unsloth-Llama-3.2-3B-Instruct-bnb-4bit-GGUF

1 reply

·

reacted to m-ric's post with ❤️ 25 days ago

Post

2376

Single most important thing to do today: 𝗴𝗼 𝘁𝗿𝘆 𝗤𝘄𝗤 𝗼𝗻 𝗛𝘂𝗴𝗴𝗶𝗻𝗴 𝗖𝗵𝗮𝘁!

👉 https://huggingface.co/chat/models/Qwen/QwQ-32B-Preview

2 replies

·

posted an update 28 days ago

Post

1174

I have this small utility: no_more_typo
It is running in the background and able to call the LLM model to update the text on the clipboard. I think it would be ideal to fix typos and syntax.
I have just added the option to use custom prompt templates to perform different tasks.

Details, code and executable:
https://github.com/csabakecskemeti/no_more_typo

https://devquasar.com/no-more-typo/

replied to their post about 1 month ago

Exactly :D
Overflow fridge will be replaced with a rack :)

reacted to cappuch's post with 😎 about 1 month ago

Post

1571

p104-100s are beasts. 8 gigs of VRAM, 12 tok/s on qwen 14b at q4, and 18 tok/s on 7b at q6. best thing - 20 euros each.

https://furry.engineer/@cappuch/113500349547803802

3 replies

·

posted an update about 1 month ago

Post

291

Repurposed my older AI workstation to a homelab server, it has received 2xV100 + 1xP40
I can reach huge 210k token context size with MegaBeam-Mistral-7B-512k-GGUF ~70+tok/s, or run Llama-3.1-Nemotron-70B-Instruct-HF-GGUF with 50k Context ~10tok/s (V100 only 40k ctx and 15tok/s).
Also able to Lora finetune with similar performace as an RTX3090.
It moved to the garage to no complaints for the noise from the family. Will move to a Rack soon :D

2 replies

·

reacted to fdaudens's post with 😎 about 1 month ago

Post

1636

🚀 @Qwen just dropped 2.5-Turbo!

1M token context (that's entire "War and Peace"!) + 4.3x faster processing speed. Same price, way more power 🔥

Check out the demo: Qwen/Qwen2.5-Turbo-1M-Demo

#QWEN

posted an update about 1 month ago

Post

1227

Some time ago, I built a predictive LLM router that routes chat requests between small and large LLM models based on prompt classification. It dynamically selects the most suitable model depending on the complexity of the user input, ensuring optimal performance while maintaining conversation context. I also fine-tuned a RoBERTa model to use with the package, but you can plug and play any classifier of your choice.

Project's homepage:
https://devquasar.com/llm-predictive-router/
Pypi:
https://pypi.org/project/llm-predictive-router/
Model:
DevQuasar/roberta-prompt_classifier-v0.1
Training data:
DevQuasar/llm_router_dataset-synth
Git:
https://github.com/csabakecskemeti/llm_predictive_router_package

Feel free to check it out, and/or contribute.

reacted to singhsidhukuldeep's post with 👍 about 1 month ago

Post

2299

Good folks at @nvidia and @Tsinghua_Uni have released LLAMA-MESH - A Revolutionary Approach to 3D Content Generation!

This innovative framework enables the direct generation of 3D meshes from natural language prompts while maintaining strong language capabilities.

Here is the Architecture & Implementation!

>> Core Components

Model Foundation
- If you haven't guessed it yet, it's built on the LLaMA-3.1-8B-Instruct base model
- Maintains original language capabilities while adding 3D generation
- Context length is set to 8,000 tokens

3D Representation Strategy
- Uses the OBJ file format for mesh representation
- Quantizes vertex coordinates into 64 discrete bins per axis
- Sorts vertices by z-y-x coordinates, from lowest to highest
- Sorts faces by the lowest vertex indices for consistency

Data Processing Pipeline
- Filters meshes to a maximum of 500 faces for computational efficiency
- Applies random rotations (0°, 90°, 180°, 270°) for data augmentation
- Generates ~125k mesh variations from 31k base meshes
- Uses Cap3D-generated captions for text descriptions

>> Training Framework

Dataset Composition
- 40% Mesh Generation tasks
- 20% Mesh Understanding tasks
- 40% General Conversation (UltraChat dataset)
- 8x training turns for generation, 4x for understanding

Training Configuration
- Deployed on 32 A100 GPUs (for Nvidia, this is literally in-house)
- 21,000 training iterations
- Global batch size: 128
- AdamW optimizer with a 1e-5 learning rate
- 30-step warmup with cosine scheduling
- Total training time: approximately 3 days (based on the paper)

This research opens exciting possibilities for intuitive 3D content creation through natural language interaction. The future of digital design is conversational!

posted an update about 1 month ago

Post

1508

I've built a small open utility pip package called LLM-Forwarder that allows you to inject context, such as adding a private RAG, into existing chat applications by forwarding the app through the LLM-Forwarder. In the forwarder server, you can configure custom code to re-process chat messages and alter the user prompt, for example, by adding extra context.

https://pypi.org/project/llm-forwarder/
More details
https://devquasar.com/llmforwarder/

Csaba Kecskemeti PRO

AI & ML interests

Recent Activity

Organizations

csabakecskemeti's activity