Impressive that this is the same model they used for the International Mathematical Olympiad a few weeks ago
MACHINE LEARNING
-
Ultra-Scale Playbook: Training DeepSeek-V3 with Advanced Parallelism
By
–
The Ultra-Scale Playbook (the large-scale LLM training guide from the @huggingface science team) is out now! It is a 246-page, very nicely designed PDF that walks you through learning how to train your own DeepSeek-V3 model using:
• 5D parallelism, • ZeRO,
• fast kernels, -
Phi-Ground Tech Report: Advancing GUI Perception
By
–
Phi-Ground Tech Report Advancing Perception in GUI Grounding
-
Seed-Prover: Deep and Broad Reasoning for Automated Theorem Proving
By
–
Seed-Prover Deep and Broad Reasoning for Automated Theorem Proving
-
Autonomous AI System Detects Lunar Craters for Landers
By
–
système IA autonome de détection de cratères lunaires pour atterrisseurs. pic.twitter.com/hXAf3QwWDT
— VISION IA (@vision_ia) 1 août 2025système IA autonome de détection de cratères lunaires pour atterrisseurs.
-
Gemini Deep Think Model Launches with IMO Gold Medal Achievement
By
–
The Gemini Deep Think model that achieved IMO gold medal 🥇 is launched! This is a general purpose model that is not only SOTA at math/proofs but also reasoning, code and many others! 🔥
— Yi Tay (@YiTayML) 1 août 2025
The exact config that achieved IMO gold with scaled up Deep Think is being made available… https://t.co/7Ol5iZQN3lThe Gemini Deep Think model that achieved IMO gold medal is launched! This is a general purpose model that is not only SOTA at math/proofs but also reasoning, code and many others! The exact config that achieved IMO gold with scaled up Deep Think is being made available
-
Context Size and System Performance: The Bias-Variance Tradeoff
By
–
Too little context, however, may cause the system to underperform due to lack of relevant signals – much like the classic bias-variance tradeoff.
→ View original post on X — @sudalairajkumar, 2025-08-01 03:23 UTC
-
Context Engineering: The New Feature Engineering in GenAI Systems
By
–
Context Engineering is the new Feature Engineering! In modern Agentic / GenAI systems, context engineering is becoming as critical as feature engineering was in the ML era. Providing too much context can cause the model to overfit to specific details, limiting generalizability.
→ View original post on X — @sudalairajkumar, 2025-08-01 03:23 UTC
-
Mixture-of-Recursions: Adaptive Computation for Language Models
By
–
What if language models could learn to "think harder" only when they need to—allocating deep computation to challenging tokens while breezing through simple ones? We're excited to have Reza tomorrow in our AI4Science community giving a talk on Mixture-of-Recursions!
-
LoRA Compatibility and Retraining Strategies for AI Models
By
–
they are compatible with LoRAs. that doesn’t mean that LoRAs are pluggable *as-is*. While we’ve found that LoRAs sometimes work well or even better plugging them as is, we’ve found better results by retraining them for the most part. The important part is the compatibility.