Open sourceLiquid AI·14d ago·3 min readSummaryListen 0:42
LFM3 登場：特化型インテリジェンスのためのあらゆるモダリティLiquid AI introduced LFM3, a multimodal model for specialized intelligence.
News Hub
What actually shipped in agent engineering, pulled from the labs, arXiv and Hacker News.
See who we follow62 articles
-mh23DVnFm2EDs4US6gJskniHIMwVY4.png&w=160)
IndustryLiquid AI·26d ago·3 min readSummary
d1-ultra: Introducing the World’s Best Decision ModelLiquid AI announced d1-ultra, described as the world's best decision model.
Open sourceLiquid AI·28d ago·3 min read
LFM2.5-24B-A2B：高速なオンデバイスエージェントのためにLiquid AI released LFM2.5-24B-A2B, a fast on-device agent model.
-oOdUn61bDRSykk4H951SoDOHzqoOe1.png&w=160)
LFM2.5-24B-A2B: Built for Fast On-Device Agents
Liquid AI released LFM2.5-24B-A2B, a model built for fast on-device agents.
Also covered byHugging Face
Open sourceLiquid AI·34d ago·3 min readSummaryListen 0:41
Open d1: Edge decision models for text, vision, and audioLiquid AI released Open d1, edge decision models for text, vision, and audio.
Open sourceLiquid AI·47d ago·3 min read
LFM2.5-VL-DSpark：エッジからその先まで、視覚言語モデルを高速化Liquid AI released LFM2.5-VL-DSpark, a vision-language model optimized for edge devices.
Open sourceLiquid AI·48d ago·3 min read
Liquid AI、Qualcomm Technologies と協力し、Snapdragon 搭載デバイスにパーソナル AI コンテキストを提供Liquid AI partnered with Qualcomm to provide personal AI context on Snapdragon devices.
Also covered byLiquid AI
ResearchLiquid AI·78d ago·3 min readSummary
Introducing Pipette: A benchmarking suite for on-device intelligenceLiquid AI introduced Pipette, a benchmarking suite for evaluating on-device intelligence systems.
Open sourceLiquid AI·82d ago·3 min read
LFM2.5-DSpark：H100 から MacBook まで最大 3.2 倍高速な推論LFM2.5-DSpark model achieves up to 3.2x faster inference from H100 GPUs to MacBook.
Open sourceLiquid AI·83d ago·3 min read
LFM2.5 Q4_0：エッジ展開のための量子化認識蒸留LFM2.5 Q4_0 uses quantization-aware distillation for edge deployment optimization.
Open sourceLiquid AI·83d ago·3 min readSummaryListen 0:44
LFM2.5 Q4_0: Quantization-Aware Distillation for Edge DeploymentLFM2.5 Q4_0 applies quantization-aware distillation for edge device deployment.
Also covered byHugging Face
Open sourceLiquid AI·90d ago·3 min read
LFM2.5-VL-3B：エッジ向けの、より高性能で高速な視覚言語モデルLFM2.5-VL-3B is a faster, higher-performance vision-language model for edge devices.
Open sourceLiquid AI·97d ago·3 min readSummaryListen 0:37
MacPaw Partners with Liquid AI to Bring On-Device AI to Millions of Mac UsersMacPaw partners with Liquid AI to deploy on-device AI models for Mac users.
Open sourceLiquid AI·98d ago·3 min read
LFM2.5-2.6B：エージェントをあらゆる場所へLFM2.5-2.6B model enables agent deployment across diverse computing environments.
Open sourceLiquid AI·105d ago·3 min read
LFM2.5-Encoders：長文脈でも、CPU でも速いLFM2.5-Encoders provide fast processing for long contexts on CPU hardware.
Also covered byHugging Face
Open sourceLiquid AI·112d ago·3 min readSummaryListen 0:47
Tokenizer Expansion: Upgrading a Model's Tokenizer in PlaceLiquid AI describes a technique to expand and upgrade a model's tokenizer without retraining.
Also covered byarXiv cs.AI
Open sourceLiquid AI·126d ago·3 min read
Final Token Preference Optimization によるドゥームループの抑制Liquid AI presents Final Token Preference Optimization method to suppress doom loops in AI systems.
Open sourceLiquid AI·126d ago·3 min read
Reducing Doom Loops with Final Token Preference OptimizationLiquid AI proposes Final Token Preference Optimization to reduce doom loops in language models.
Open sourceLiquid AI·138d ago·3 min read
LFM2.5-230M：どこでも動作するように作られたモデルLiquid AI releases LFM2.5-230M model designed to run anywhere.
Open sourceLiquid AI·138d ago·3 min read
LFM2.5-230M: Built to Run AnywhereLiquid AI released LFM2.5-230M, a model designed to run on diverse hardware platforms.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.