Hacker News usually provides an outbound discussion, not an original image.
ElementaryUI – Swift, in the Browser
Original source: elementary.codes
by frizlab · 118 points · 35 comments, no article summary provided
20 signalsFocus: Dev / Open Source, AI Models
Hacker News usually provides an outbound discussion, not an original image.
Original source: alpinelinux.org
by sashk · 254 points · 122 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
IT之家 10 月 30 日消息,小米官方今日宣布开源轻量级文档理解模型 Xiaomi-OCR-0,模型权重与推理代码已同步上线 Hugging Face 和 GitHub。 官方给出的参数规模为 0.8B,定位是单一模型同时覆盖文档解析、关键信息抽取与 OCR 视觉问答三类任务。 在文档解析方向,该模型可识别正文、表格与数学公式,按阅读顺序输出结构化结果,原生支持 Markdown、LaTeX 格式。关键信息抽取面向发票、收据、卡证和业务表单,依据指定字段定位并输出 JSON 等结构化数据,可接入自动化处理流程。 OCR 视觉问答则允许围绕文档与图表内容进行跨区域检索、数值计算和语义推理,而非仅做文字转写。小米方面表示,这一组合覆盖了从底层文字感知到高层内容理解的链路。 据小米官方披露,Xiaomi-OCR-0 基于 Qwen3.5-0.8B-Base 基座构建,训练累计使用约 1.7 亿条 OCR 与文档理解样本。数据侧采用异构专家共识机制,对正文、表格、公式等不同区域由多个 OCR 专家模型交叉校验并投票筛选高置信标注。 针对复杂表格与公式,团队引入“预测 — 重渲染 — 视觉比对”的闭环校验:将候选解析结果重新渲染为图像,与原始文档区域做像素和结构级比对,定位行列错位或符号缺失后定向修正。此外还按版面复杂度、字体形态与物理退化因素合成困难样本,补充多语种、古籍手写、屏幕摩尔纹与扭曲光照等长尾数据。 训练分为三个阶段:Q-Mask 文本锚定强化视觉局部区域与文字 Token 的空间对应;OCR 持续预训练以百亿级 Token 的多任务混合数据打底;混合任务强化学习则在困难样本上使用结构编辑距离、表格匹配度和问答准确率作为可验证奖励信号。 小米公布的评测结果显示,在模拟扫描、纸张弯曲、屏幕翻拍、光照变化和倾斜拍摄五种真实采集方式的 Real5-OmniDocBench 上,Xiaomi-OCR-0 得分为 95.24 分,在其对比模型中排名第一。在 Wild-OmniDocBench 上得分为 87.94 分。
Hacker News usually provides an outbound discussion, not an original image.
Original source: githubstatus.com
by pelagicAustral · 6 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
by popcar2 · 4 points · 5 comments · Hi HN! I just released a new minimalist website that lets you play over 150 PICO 8 games on the web that I curated myself. It's fully open source and made from the ground up to be tiny and fast, allowing you to switch games instantly by hotswapping them into the game player. The
Hacker News usually provides an outbound discussion, not an original image.
Original source: github.com
by modinfo · 10 points · 0 comments, no article summary provided
36Kr news flashes usually do not include images.
36氪获悉,近日,记忆张量(MemTensor)发布消息称,其记忆原生模型MemT 1.0自10月20日上线Hugging Face、GitHub以来,累计下载量已突破10万次,日均下载超1万次。
Hacker News usually provides an outbound discussion, not an original image.
Original source: twitter.com
by nreece · 6 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: arstechnica.com
by cdrnsf · 10 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: duckdb.org
by andreasha · 17 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: pypi.org
by karpetrosyan · 15 points · 2 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: github.com
by luxshan_t · 5 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: theregister.com
by theanonymousone · 6 points · 2 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: github.com
by akashpatel7204 · 12 points · 4 comments, no article summary provided
Original source: github.com
by g0xA52A2A · 81 points · 56 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: thenewstack.io
by theanonymousone · 47 points · 22 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
Original source: techcommunity.microsoft.com
by clairegiordano · 4 points · 0 comments, no article summary provided
Hacker News usually provides an outbound discussion, not an original image.
by TekMol · 25 points · 3 comments · The prompt issue: https://github.com/anthropics/claude-code/issues/81138 The LLM sends it to websites: https://www.reddit.com/r/ClaudeCode/comments/1xhqvzn/opus_55_just_shared_my_email_address/
This source did not provide an image.
当 Claude 开始承担大量编码工作,Anthropic 的工程师正在转向哪些更难的问题?
Hacker News usually provides an outbound discussion, not an original image.
by soasme · 8 points · 6 comments · I burned about $500 for this fun project: Small Python. Initial PoC used 3 resets of Codex in one session but all later development happens in Claude Code, also in one Opus 5.5 session over three days. The outcome is a quickjs-like implementation for Python. Almost full Python 3.