Loading…
Topic
617 posts about Machine Learning. Every summary links to the original.
See the idea
A walkthrough of how a 3D CT model turns a voxel volume into a structured diagnostic report: isotropic resampling, 3D patch tokens, spatial encoding, and language-model reasoning.
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Newest first
Training and aligning large language models and small language models requires datasets tailored for complex reasoning, preference tuning, multi-turn questioning, and format conversion. To eliminate the need for bespoke data preparation scripts, SyGra provides a low-code and no-code Python framework for automated dataset creation, transformation, and alignment. The library integrates with diverse inference backends, including vLLM, Hugging Face TGI, Triton, and Ollama, allowing teams to focus on prompt engineering rather than pipeline infrastructure. It supports data workflows ranging from converting PDFs and knowledge bases into question-answering pairs to filtering low-quality samples and generating Direct Preference Optimization datasets. These plug-and-play workflows reduce manual curation effort while accelerating model fine-tuning and retrieval-augmented generation pipelines.
At MongoDB.local NYC 2025, MongoDB introduced new capabilities and product updates tailored for artificial intelligence workloads. MongoDB 8.2 was unveiled alongside public previews of full-text and vector search for both MongoDB Community Edition and Enterprise Server. The company also detailed its integration with Voyage AI embedding and reranking models to deliver precision retrieval across raw data, metadata, and vectors. To address legacy infrastructure migration hurdles, MongoDB launched the Application Modernization Platform, which combines AI-assisted tooling and specialized workflows to accelerate legacy application refactoring two to three times faster. These architectural updates aim to position the JSON document model as the persistent memory and context store required by autonomous, agentic AI workflows.
MongoDB unveiled several core database updates aimed at supporting artificial intelligence workloads during MongoDB.local NYC 2025. The release of MongoDB 8.2 introduces performance enhancements alongside Voyage AI embedding models and rerankers for AI applications. Furthermore, Search and Vector Search are now available in public preview for both MongoDB Community Edition and MongoDB Enterprise Server. To address the costs and operational friction of legacy systems, MongoDB launched the Application Modernization Platform (AMP), which combines AI-powered tools with specialized talent to accelerate database migrations. These capabilities position the JSON-based document model to supply the persistent state, memory, and retrieval mechanisms required by emerging agentic AI architectures.
Cloud Security Alliance and Noma Security introduced RiskRubric.ai to provide standardized, transparent risk assessments across the open AI model ecosystem. The framework evaluates AI models across six pillars—transparency, reliability, security, privacy, safety, and reputation—using over 1,000 reliability tests, 200 adversarial security probes, automated code scanning, and harmful content evaluations. Each model receives 0–100 scores and A–F letter grades, supplemented by specific vulnerability findings and recommended mitigation strategies to assist deployment filtering. Initial benchmark results across models showed composite scores ranging from 47 to 94 with a median of 81, revealing polarized safety distributions and indicating that security hardening directly correlates with reduced safety risks.
Hugging Face has integrated Public AI as a supported Inference Provider on the Hugging Face Hub. Public AI operates as a nonprofit, open-source project providing access to sovereign and public models from institutions such as the Swiss AI Initiative and AI Singapore. Its distributed infrastructure combines a vLLM-powered backend serving OpenAI-compatible APIs across partner-donated clusters with a global load-balancing routing layer. Users can access these models through the Hugging Face web UI, Python client SDK, and JavaScript SDK using either direct provider API keys or routed Hugging Face tokens. At the time of announcement, inference through the Public AI provider is free of charge, supported by donated GPU time and advertising subsidies.
LeRobotDataset:v3.0 is a new release for lerobot, Hugging Face’s robotics library, addressing file-system limitations encountered when LeRobotDataset:v2 stored one episode per file at million-episode scale. Version 3 packs multiple episodes into shared Parquet and MP4 files, while relational JSON and chunked episode metadata identify episode boundaries and locate the corresponding tabular and visual data. The format separates tabular data, visual data, and metadata; it stores low-dimensional readings in Apache Parquet, concatenates camera frames into MP4 videos, and supports native temporal windowing through delta_timestamps. It also introduces StreamingLeRobotDataset for processing batches from the Hugging Face Hub without downloading the full dataset locally, and provides a one-line conversion utility for datasets in the LeRobotDataset format. The release is planned for lerobot-v0.4.0 and is presented as a step toward datasets with millions of episodes.
Generative artificial intelligence tools produce images, video, audio, and text that have become increasingly difficult to distinguish from real-world media captures. To improve synthetic media transparency, Hugging Face introduced visible watermarking capabilities directly into the Gradio web application framework. Developers can now overlay visual watermarks on image and video outputs by specifying a single watermark parameter using file paths, open image objects, or NumPy arrays. The implementation also supports QR code watermarks, which can match the visual style of generated media while conveying detailed background information. Furthermore, the Gradio Chatbot component accepts a text watermark parameter that automatically attaches attribution metadata whenever users copy generated responses to their clipboards.
WRITER released three open models in the Palmyra-mini family ranging between 1.5B and 1.7B parameters, designed for efficient inference and specialized reasoning tasks. Built on the Qwen architecture, the release includes the standard palmyra-mini base model alongside two Chain of Thought variants, palmyra-mini-thinking-a and palmyra-mini-thinking-b. In benchmark evaluations, palmyra-mini achieved 52.6% on Big Bench Hard, while palmyra-mini-thinking-a reached 82.87% on GSM8K and palmyra-mini-thinking-b reached 92.5% on AMC23. For palmyra-mini-thinking-b, applying reinforcement learning fine-tuning to an OpenReasoning-Nemotron-1.5B base improved single-shot pass@1 accuracy while reducing sampling diversity and majority@64 performance. The models are available in GGUF and MLX-BF16 quantizations and support inference engines including vLLM, SGLang, TRTLLM, and TGI.
To support OpenAI's GPT-OSS series of models, the transformers library introduced several performance upgrades that apply across supported architectures. The release integrates zero-build custom kernels downloadable directly from the Hub, reducing external dependency bloat and compilation friction for operations like Liger RMSNorm and MegaBlocks MoE. Native support for MXFP4 quantization groups vector elements into 32-value blocks with shared scales, allowing GPT-OSS 20B to fit in roughly 16 GB of VRAM and GPT-OSS 120B in roughly 80 GB. In addition, transformers incorporates Flash Attention 3 with attention sinks, continuous batching via the generate_batch API for experimentation, and automatic memory pre-allocation to speed up model loading on GPUs.
Together AI and Hugging Face have introduced an integration enabling developers to fine-tune compatible Hugging Face Hub models directly on Together AI's managed infrastructure. To launch a fine-tuning job via the Python SDK, users provide a base model from Together's catalog as a configuration template alongside the target Hugging Face repository identifier. This base model template dictates GPU allocation, memory configuration, training pipelines, and inference setup for custom models with matching architectures and sizes. The workflow operates bidirectionally, pulling from public or token-authenticated private repositories and optionally pushing completed checkpoints back to the Hub upon completion. Geared toward CausalLM models under 100 billion parameters, the capability enables faster iteration cycles and domain adaptation without custom DevOps infrastructure.
Small language models often struggle to compete with large frontier models on complex, agentic data science tasks. To improve notebook-based reasoning, researchers simplified agent scaffolding down to roughly two hundred lines of code with dedicated execution and final answer tools, boosting baseline easy accuracy on the DABStep benchmark from 44.4 percent to 59.7 percent. They constructed a curated training dataset by deduplicating two terabytes of Kaggle notebooks, automatically fetching five terabytes of linked datasets, and scoring educational value and relevance using Qwen3-32B. Question-answer pairs grounded in verified execution traces were generated to fine-tune compact Qwen3-4B thinking and instruct models. The team released the trained models alongside the dataset and execution sandboxes, establishing a foundation for reinforcement learning and distillation on notebook workflows.
mmBERT is a massively multilingual encoder model trained on more than 3T tokens across over 1,800 languages to improve upon existing multilingual architectures like XLM-R. Built upon ModernBERT with a Gemma 2 tokenizer, mmBERT employs a three-phase training curriculum consisting of pre-training on 60 languages, mid-training on 110 languages, and a final decay phase covering 1,833 languages. The training pipeline integrates an inverse mask ratio schedule, dynamic language temperature annealing, and TIES merging across three decay variants. Benchmark evaluations demonstrate strong natural language understanding on English GLUE and multilingual XTREME, as well as competitive retrieval performance on MTEB v2 and CoIR. The release includes standard base and small models alongside open training data and checkpoints.
Google DeepMind released EmbeddingGemma, a multilingual embedding model with 308 million parameters and a 2048-token context window designed for on-device applications. Based on the Gemma3 transformer backbone, the architecture replaces causal attention with bidirectional attention to function as an encoder, followed by mean pooling and two dense layers producing 768-dimensional vectors. The model incorporates Matryoshka Representation Learning, allowing outputs to be truncated down to 512, 256, or 128 dimensions for reduced memory and storage footprints. Trained on approximately 320 billion multilingual tokens across more than 100 languages, the quantized model operates under 200 MB of RAM. In domain-specific evaluations on the MIRIAD dataset, fine-tuning increased NDCG@10 from 0.8340 to 0.8862, outperforming larger baselines.
SandboxAQ released the Structurally Augmented IC50 Repository (SAIR), an open-source dataset containing 5.24 million computationally co-folded 3D protein-ligand structures paired with empirical IC50 binding potency data. The repository addresses training data scarcity in structure-based drug discovery, where over 40 percent of the included target proteins lack experimental structures in the Protein Data Bank. To construct the dataset, engineers executed over 130,000 GPU hours of the Boltz1 co-folding model on 760 NVIDIA H100 processors hosted via NVIDIA DGX Cloud on Google Cloud Platform. Infrastructure optimizations maintained over 95 percent GPU compute utilization, compressing the generation timeline from three months to three weeks. Quality validation using PoseBusters confirmed that 97 percent of the predicted complexes met physical plausibility and chemical sanity standards.
Shopify’s Sidekick evolved from a simple tool-calling system into an agentic assistant that helps merchants manage stores through natural language, exposing scaling challenges in architecture, evaluation, and training. As its inventory grew beyond 50 tools, the system prompt accumulated conflicting special cases and became difficult to maintain; JIT instructions instead return relevant guidance with tool data. For evaluation, the team uses Ground Truth Sets based on production distributions, human labels from at least three product experts, statistical agreement measures, specialized LLM judges calibrated against humans, and merchant simulation. Judge correlation rose from Cohen’s Kappa 0.02 to 0.61 against a human baseline of 0.69, while syntax validation improved from about 93% to 99% after reward-hacking fixes. GRPO training exposed opt-out, tag, and schema hacking, and the post recommends modularity, procedural plus semantic validation, user simulation, and iterative judge refinement for production agentic systems.
Grab describes how it evolved its machine learning feature store as increasingly complex models began requiring high-cardinality keys and high-dimensional data. It replaced its initial Amphawa approach with feature tables: data-lake tables whose contents are atomically ingested into serving storage, with emphasis on consistency and reproducibility. The workflow stores feature tables and version-control metadata as Parquet objects in Amazon S3, validates schemas, loads data through reverse ETL into Aurora Postgres, and warms new tables through shadow reads before switching. The redesign supports contextual retrieval, atomic collection-wide updates, and isolation between reading and writing capacity; its SDK remains a thin database-driver wrapper that translates queries to SQL. Grab also chose decentralised deployments for version flexibility, rollout strategies, noisy-neighbor mitigation, and improved caching, concluding that Amazon Aurora for Postgres provides a robust, scalable, cost-effective foundation.
Shopify’s Global Catalogue addresses fragmented, multilingual, multimodal product data that makes semantic search, faceting, matching, and AI-driven discovery difficult across millions of merchants. The system combines product data, understanding, matching, and reconciliation layers to standardize metadata, identify duplicate listings, and create canonical records with Universal Product IDs. Shopify fine-tunes vision large language models for multiple tasks per catalogue entity, while streaming infrastructure processes more than 10 million product updates daily and serves 40 million multimodal LLM-powered inferences. The resulting data supports merchant-admin suggestions, search, recommendations, embeddings, and conversational commerce, although scalability, accuracy, latency, model consolidation, graph reasoning, and continuous retraining remain active challenges.
Shopify frames commerce as a complex, continuous optimization problem in which merchants have different definitions of success and need support across the business lifecycle. Its machine-learning work uses a historical dataset covering millions of businesses to optimize product metadata, transaction fraud assessment, merchant assistance, product discovery, capital decisions, behavioral modeling, and search query rewriting. The systems include fine-tuned Qwen multimodal models, LLaMa models and MCPs for Sidekick, Nomic embeddings for billions of products, a tabular transformer for forecasting merchant GMV, and experiments with HSTU sequence models. To support this work, Shopify partners with GCP, Nebius, CentML, and Toloka, while providing access to GPUs, data services, frontier models, and internal tooling; the post also presents these efforts as part of its hiring pitch.
React Native WebGPU and React Native Skia are presented as a graphics stack for advanced 2D, 3D, and general-purpose GPU workloads in native applications. The work uses Skia Graphite with Dawn, Google’s open-source WebGPU implementation, to replace separate OpenGL and Metal backend implementations with a unified architecture, while enabling Three.js, React Three Fiber, TensorFlow.js, and compute shaders. Three.js Shading Language lets the same shaders target WebGPU and WebGL, and Shopify’s Black Friday globe prototypes run natively with identical results to the web version. Architectural changes in React Native Skia moved from a mutable SkiaDOM to an immutable fabric reconciler, producing reported animation gains of up to 50% on iOS and nearly 200% on Android, plus 98% of reported Android crashes fixed. Graphite remains experimental, but adds dedicated-thread canvases and bidirectional 2D/3D composition, while Skia also supports macOS, tvOS, and headless Node.js rendering.
Matt Greenwood, Chief Innovation Officer at Two Sigma, outlines strategies for managing technological change, integrating artificial intelligence into systematic investment, and building supportive engineering organizations. Greenwood describes sustained innovation through the S-curve using an epsilon and omega approach, combining small iterative steps with a broad long-term vision. To direct resources amid rapid advances in machine learning and large language models, he presents a functional framework categorizing AI roles into advisory insights, oracle outcome validation, operational task automation, and agentic coordination. This categorization aims to automate routine workflows while keeping human creativity, control, and higher-level thinking at the center of the investment process. Additionally, the organization fosters employee engagement through initiatives such as an internal hacker lab where cross-disciplinary teams build projects ranging from robots to racing simulators.
The best engineering reads, once a week.
