Open sourceHugging Face·1d ago·3 min readSummaryListen 0:49
Accelerating vision-language models with LFM2.5-VL-DSparkHugging Face released LFM2.5-VL-DSpark for accelerating vision-language models.
News Hub
What actually shipped in agent engineering, pulled from the labs, arXiv and Hacker News.
See who we follow556 articles
Open sourceHugging Face·2d ago·3 min readSummaryListen 0:53
**Know Who Spoke When: Build Real-Time, Multi-Speaker AI with NVIDIA Nemotron 3 Diarization**NVIDIA Nemotron 3 Diarization enables real-time multi-speaker identification for AI applications.
Also covered byBaseten
ResearchHugging Face·4d ago·3 min readSummaryListen 0:47
How UK AISI and EvalEval Are Making Benchmark Results ReproducibleUK AISI and EvalEval work to make AI benchmark results reproducible.
Open sourceHugging Face·4d ago·3 min readSummaryListen 0:41
Transformers now runs llama.cpp quantsHugging Face Transformers library now supports running llama.cpp quantized models.
Open sourceHugging Face·4d ago·3 min readSummaryListen 0:51
Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization ProblemLLM pruning can be formulated as an Ising optimization problem for block removal.
Open sourceHugging Face·5d ago·3 min readSummaryListen 0:44
tokenizers v1: encode, decode and scaling, measuredTokenizers v1 measures encode, decode performance and scaling characteristics.
Open sourceHugging Face·10d ago·3 min readSummaryListen 0:53
Your Agent Aced the Task. Will It Do It Again?Article examines whether AI agents consistently reproduce successful task performance.
Open sourceHugging Face·16d ago·3 min readSummaryListen 0:46
Rebuilding AUTOMATIC1111 with Gradio WorkflowAUTOMATIC1111 is being rebuilt using Gradio Workflow.
Open sourceHugging Face·16d ago·3 min readSummaryListen 0:34
Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCLAsync GRPO with LoRA training runs across Hugging Face Jobs using a bucket and proxy without NCCL.
ResearchHugging Face·16d ago·3 min readSummaryListen 0:52
IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly licenseIBM released Granite Time Series PatchTST-FM-r2 model with commercial-friendly license.
Open sourceHugging Face·22d ago·3 min readSummaryListen 0:45
NeoMME: an efficient Multimodal-native and Multilingual EncoderNeoMME is an efficient multimodal-native and multilingual encoder model.
Open sourceHugging Face·23d ago·3 min readSummaryListen 0:43
Give Your Coding Agents a Memory You OwnCoding agents can be equipped with persistent memory systems owned and controlled by users.
Open sourceHugging Face·23d ago·3 min readSummaryListen 0:42
Training a coding model to paint watercolours with TRL and OpenEnvA coding model was trained to generate watercolor paintings using TRL and OpenEnv frameworks.
Open sourceHugging Face·23d ago·3 min readSummaryListen 0:42
Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO StepsA 350M parameter model was fine-tuned using 100 GRPO steps to improve structured output generation.
IndustryHugging Face·25d ago·3 min readSummaryListen 0:45
Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AIHugging Face released 200+ WebGPU kernels for running AI models locally.
Open sourceHugging Face·31d ago·3 min readSummary
Training and Finetuning Multi-Vector Embedding Models with Sentence TransformersSentence Transformers enables training and finetuning of multi-vector embedding models.
Open sourceHugging Face·31d ago·3 min readSummaryListen 0:39
Granite 4.2 LLMs: How They're BuiltGranite 4.2 LLMs are large language models with documented construction methods.
Open sourceHugging Face·31d ago·3 min readSummaryListen 0:42
Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision originalA 4-bit quantized model achieves better performance than its original full-precision version through quantization-aware healing techniques.
ResearchHugging Face·36d ago·3 min readSummaryListen 0:42
How Hugging Face Inference Endpoints, Jobs, and Buckets Power Search on Papers with CodeHugging Face Inference Endpoints, Jobs, and Buckets provide infrastructure for the Papers with Code search functionality.
ResearchHugging Face·36d ago·3 min readSummaryListen 0:55
Measuring benchmark optimization in speech recognitionMeasuring benchmark optimization in speech recognition.
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.