Tag
TypeLLM 开源发布，通过 JSON Schema 将整个 LLM 工作流建模为依赖图，并基于 SGLang 的约束解码在单次 API 调用中执行，实现类型安全、schema 保证的输出，无需修改 Qwen 等模型权重。
AI Monster Brawl sent the same prompt to all 325 OpenRouter text models to build a monster fighter within a 1000-point budget; a third failed on the first try, mostly by overspending, and 7% never produced a valid fighter. The post analyzes common failure modes like budget overruns, invalid parts, and malformed JSON.
本文提出 FinVector-Market-4B，使用 rank-16 LoRA 在 22,000 条语料上对 Qwen3.5-4B 进行结构化金融任务的适配研究，表明在匹配提示契约下，紧凑型领域适配能带来超出输出格式学习的显著任务收益。
Chargebee launched AUDR (Agent Usage Detail Record), an open standard defining a common JSON schema for capturing who initiated an agent run and what it cost across every system the run touches, inspired by telecom's Call Detail Record.
A high-throughput inference engine for structured information extraction on Apple Silicon using MLX, offering parallel constrained decoding with 5.6x to 7.0x latency reductions and 100% schema validity.
AI Engineer shares a data library and MCP integration for AI agents, offering access to public collections, tools, and schemas without requiring sign-in.
The paper introduces OrderBench, a benchmark for restaurant ordering LLM agents that evaluates semantic reliability beyond schema validity, demonstrating that structured output modes can achieve perfect schema validity while still having high semantic error rates.
Object Aligner is an open-source Python library that deterministically scores two JSON objects by recursively aligning their trees, using Hungarian algorithm for unordered collections and sequence alignment for ordered ones. It introduces referential alignment for graphs/hypergraphs and can be used as a reward function in LLM prompt optimization.
GolemUI is a declarative form engine that defines forms as JSON schemas, offering accessibility, i18n, and multi-framework support out of the box.
Datalab releases lift, a model that extracts structured JSON from PDFs and images using schema-constrained decoding, with local and hosted inference options.
This article explains how to use GRPO to fine-tune an LLM (Qwen3-8B) for reliable JSON structured output, improving schema accuracy from 62% to 82%, surpassing GPT-4.1's 58%.
A developer shares lessons from building a local document-to-JSON extractor using llama3.2 3B on Ollama, highlighting that deterministic post-processing and schema-constrained outputs matter more than model size, while seeking feedback on hallucination and context truncation issues with long documents.
Open Envelope provides an open JSON Schema for defining AI agent teams with roles, handoffs, human checkpoints, and access policies, aiming for portability across frameworks.
The author shares findings on the lack of portability for JSON Schema structured outputs across AI providers like OpenAI, Gemini, and Anthropic, highlighting inconsistencies in constraint enforcement and offering practical advice for robust integration.