All tags
Company: "flashattention"
not much happened today
eagle-3.1 unigram-tokenizer qwen-3.5 deepseek-v4-pro mimo deep-agents-v0.6 397b-parameter-model eaglecorp vllm_project perplexity_ai alibaba lightseek nvidia mooncake flashattention kimmonismus deepseek xiaomi langchain baseten trajectory clay harvey decagon mercor rogo rlm inference-optimization long-context speculative-decoding tokenization attention-mechanisms kv-cache cache-hierarchy agent-engineering model-harness-memory-fit continual-learning quantization autoscaling memory-centric-agents evaluation-automation kimmonismus _luofuli vtrivedy10
Inference optimization is increasingly architectural, with EAGLE 3.1 improving speculative decoding and long-context handling, collaborating with vLLM and TorchSpec. Perplexity open-sourced a rebuilt Unigram tokenizer cutting CPU use by 5–6× and achieving 63 µs at 514 tokens. Qwen3.5 hits 580 tokens/s via joint efforts from Alibaba, LightSeek, NVIDIA, Mooncake, and FlashAttention-4 contributors. Price cuts in APIs from Chinese labs are sustainable due to structural KV-cache and attention improvements, exemplified by DeepSeek V4-Pro and Xiaomi MiMo reducing caching costs significantly.
Agent engineering shifts focus from model quality to model-harness-memory fit, with LangChain releasing Deep Agents v0.6 and tools like LangSmith Engine automating evaluation loops. Trajectory launched a continual learning platform with $15M funding and partners like Clay and Harvey, supporting large models including a 397B-parameter model deployed on autoscaled H100 infrastructure. Open-source memory-centric agents and minimal training harnesses also gained attention.
Microsoft AgentInstruct + Orca 3
mistral-7b orca-2.5 microsoft-research apple tencent hugging-face synthetic-data fine-tuning instruction-following transformers model-performance hallucination-detection dataset-quality flashattention mixture-of-experts philschmid sama bindureddy rohanpaul_ai zachtratar dair_ai
Microsoft Research released AgentInstruct, the third paper in its Orca series, introducing a generative teaching pipeline that produces 25.8 million synthetic instructions to fine-tune mistral-7b, achieving significant performance gains: +40% AGIEval, +19% MMLU, +54% GSM8K, +38% BBH, +45% AlpacaEval, and a 31.34% reduction in hallucinations. This synthetic data approach follows the success of FineWeb and Apple's Rephrasing research in improving dataset quality. Additionally, Tencent claims to have generated 1 billion diverse personas for synthetic data. On AI Twitter, notable discussions included a shooting incident at a Trump rally and recent ML research highlights such as FlashAttention-3, RankRAG, and Mixture of A Million Experts.
We Solved Hallucinations
gpt-2 flashattention-3 lynx meta-ai-fair nvidia princeton colfax patronus-ai databricks mosaic-ai openai compute-hardware gpu-optimization flashattention llm-evaluation hallucination-detection vision benchmarking synthetic-data model-training karpathy tri_dao giffmana vikhyatk dbrxmosaicai
Reddit's URL structure causes link errors in AI-generated summaries, especially with NSFW content affecting models like Claude and GPT-4. The team fixed this glitch while still leveraging LLMs for summarizing Reddit content. GPT-2 training costs have dramatically dropped to ~$672 using H100 GPUs and software improvements like CUDA and FlashAttention. FlashAttention-3 was released, achieving up to 740 TFLOPS on H100 GPUs, with FP8 nearing 1.2 PFLOPS, developed collaboratively by Meta, NVIDIA, Princeton, and Colfax. Hopper GPUs enable major speedups with new hardware features. Synthetic data may not improve vision tasks, as shown in recent research. The Avocado360 benchmark evaluates vision-language models' ability to detect avocados in images. Lynx, a hallucination detection model for LLMs, was introduced for real-world healthcare and fintech applications, trained by Patronus AI on Databricks Mosaic AI using Composer.