All tags
Topic: "training-efficiency"
Gemini's AlphaEvolve agent uses Gemini 2.0 to find new Math and cuts Gemini cost 1% — without RL
gemini gpt-4.1 gpt-4o-mini o3 o4-mini google-deepmind openai algorithm-discovery coding-agents matrix-multiplication optimization reinforcement-learning model-weights training-efficiency safety-evaluations instruction-following coding-tasks model-releases _philschmid scott_swingle alex_dimakis henry jason_wei kevinweil michpokrass scaling01 gdb
Deepmind's AlphaEvolve, a 2025 update to AlphaTensor and FunSearch, is a Gemini-powered coding agent for algorithm discovery that designs faster matrix multiplication algorithms, solves open math problems, and improves data center and AI training efficiency. It achieves a 23% faster kernel speedup in Gemini training and surpasses state-of-the-art on 20% of applied problems, including improvements on the Minimum Overlap Problem and Kissing number problem. Unlike Deep-RL, it optimizes code pieces rather than model weights. Meanwhile, OpenAI released GPT-4.1 in ChatGPT, specializing in coding and instruction following, with a faster alternative GPT-4.1 mini replacing GPT-4o mini for all users. OpenAI also launched the Safety Evaluations Hub and the OpenAI to Z Challenge using o3/o4 mini and GPT-4.1 models to discover archaeological sites. "Maybe midtrain + good search is all you need for AI for scientific innovation" - Jason Wei.
not much happened today
vllm deepseek-v3 llamaindex openai deepseek qdrant twilio llamaindex elevenlabs training-efficiency parallelism cpu-offloading gradient-descent mixture-of-experts fp8-precision memory-optimization ai-voice-assistants coding-assistants document-processing version-control learning-rate-schedules federated-learning agentic-systems multi-agent-systems deliberative-alignment chain-of-thought on-device-ai multimodality francois-fleuret daniel-hanchen aaron-defazio fchollet elad-gil wojciech-zaremba richard-socher
ChatGPT, Sora, and the OpenAI API experienced a >5 hour outage but are now restored. Updates to vLLM enable DeepSeek-V3 to run with enhanced parallelism and CPU offloading, improving model deployment flexibility. Discussions on gradient descent in top-k routing MoE and adoption of FP8 precision focus on training efficiency and memory optimization. AIDE, an AI voice medical assistant by Team Therasync, leverages Qdrant, OpenAI, and Twilio. DeepSeek-Engineer offers AI-powered coding assistance with structured outputs. LlamaIndex integrates LlamaCloud and ElevenLabs for large-scale document processing and voice interaction. Insights on version control with ghstack and advocacy for linear decay learning rate schedules highlight best practices in AI development. Experts predict smaller, tighter models, true multimodal models, and on-device AI in 2025. Proposals for planetary-scale federated learning and community AGI moonshots emphasize future AI directions. Discussions on agentic systems, multi-agent workflows, and deliberative alignment through chain of thought reasoning underscore AI safety and alignment efforts.
Mamba-2: State Space Duality
mamba-2 mamba transformer++ llama-3-70b gpt-3 hugging-face state-space-models perplexity training-efficiency data-pruning benchmarking multimodality video-analysis _albertgu tri_dao arankomatsuzaki _akhaliq clementdelangue karpathy
Mamba-2, a new state space model (SSM), outperforms previous models like Mamba and Transformer++ in perplexity and wall-clock time, featuring 8x larger states and 50% faster training. It introduces the concept of state space duality (SSD) connecting SSMs and linear attention. The FineWeb-Edu dataset, a high-quality subset of the 15 trillion token FineWeb dataset, filtered using llama-3-70b for educational quality, enables better and faster LLM learning, potentially reducing tokens needed to surpass GPT-3 performance. Additionally, perplexity-based data pruning using a 125M parameter model improves downstream performance and reduces pretraining steps by up to 1.45x. The Video-MME benchmark evaluates multi-modal LLMs on video analysis across multiple visual domains and video lengths.