Check out the NinjaChat AI platform over here : https://www.ninjachat.ai/
USE COUPON CODE "KING25" for 25% OFF on ALL MEMBERSHIPS ON ninjachat.ai
In this video, I'll be telling you about OpenThinker which is a new reasoning model that claims to beat the Deepseek-R1. Today, I'll test it and we'll see that how well it performs.
-----
Key Takeaways:
😃 OpenThinker is a fully open model with available model weights, datasets, and training codes.
🚀 It showcases that scaling data, verifying reasoning traces, and enlarging model size can generate state-of-the-art reasoning models.
📊 Benchmarks indicate competitive performance compared to models trained on more tokens, especially the DeepSeek R1 distill model.
💡 Two model sizes are offered (7b and 32b), with the 32b variant demonstrating superior performance on several tasks.
📝 The model uses curated datasets refined through verification, highlighting the importance of data quality.
💻 OpenThinker is accessible on Ollama and can be run locally on high-end hardware like an RTX 4090.
🤖 Despite weak coding performance, it stands out as a proof of concept for open data and efficient model distillation.
-----
Timestamps:
00:00 - Introduction
02:47 - NinjaChat (Sponsor)
03:55 - Testing
07:28 - Final Charts & Thoughts
09:18 - Ending
