https://devotd.net/en/news/github-slack-teams-multi-repository-conversations/devotd.en2026-10-18T15:50:07ZGitHub in Slack and Teams: Multi-repository support, improved conversation contexthttps://devotd.net/ja/news/github-slack-teams-multi-repository-conversations/devotd.ja2026-10-18T15:50:07ZGitHub in Slack and Teams：複数リポジトリ対応、会話の文脈活用を改善https://devotd.net/ko/news/github-slack-teams-multi-repository-conversations/devotd.ko2026-10-18T15:50:07ZGitHub in Slack and Teams: 다중 저장소 지원, 대화 맥락 활용 개선https://devotd.net/zh-hans/news/github-slack-teams-multi-repository-conversations/devotd.zh-cn2026-10-18T15:50:07ZGitHub in Slack and Teams：支持多仓库，改进对话上下文利用https://devotd.net/en/news/trl-1-16-aws-trainium-asyncgrpo-tree-attention/devotd.en2026-10-18T15:36:40ZHugging Face TRL v1.16.0: AWS Trainium training support, AsyncGRPO tree attentionhttps://devotd.net/ja/news/trl-1-16-aws-trainium-asyncgrpo-tree-attention/devotd.ja2026-10-18T15:36:40ZHugging Face TRL v1.16.0：AWS Trainiumでの学習対応、AsyncGRPOのツリーアテンションhttps://devotd.net/ko/news/trl-1-16-aws-trainium-asyncgrpo-tree-attention/devotd.ko2026-10-18T15:36:40ZHugging Face TRL v1.16.0: AWS Trainium 학습 지원, AsyncGRPO 트리 어텐션https://devotd.net/zh-hans/news/trl-1-16-aws-trainium-asyncgrpo-tree-attention/devotd.zh-cn2026-10-18T15:36:40ZHugging Face TRL v1.16.0：支持 AWS Trainium 训练，AsyncGRPO 树注意力https://devotd.net/en/news/github-copilot-hydrafusion-max-model-pool-cascade-critique/devotd.en2026-10-18T15:20:46ZGitHub Copilot HydraFusion: Stage-specific reasoning, new Max optionhttps://devotd.net/ja/news/github-copilot-hydrafusion-max-model-pool-cascade-critique/devotd.ja2026-10-18T15:20:46ZGitHub Copilot HydraFusion：段階別の推論調整、Maxの選択肢を追加https://devotd.net/ko/news/github-copilot-hydrafusion-max-model-pool-cascade-critique/devotd.ko2026-10-18T15:20:46ZGitHub Copilot HydraFusion: 단계별 추론 조정, Max 선택지 추가https://devotd.net/zh-hans/news/github-copilot-hydrafusion-max-model-pool-cascade-critique/devotd.zh-cn2026-10-18T15:20:46ZGitHub Copilot HydraFusion：分阶段调整推理强度，新增 Max 选项https://devotd.net/en/news/github-agentic-autofix-code-scanning-ga-assign-to-copilot/devotd.en2026-10-18T13:36:19ZGitHub Agentic autofix: General availability, integrated alert fixes and verificationhttps://devotd.net/ja/news/github-agentic-autofix-code-scanning-ga-assign-to-copilot/devotd.ja2026-10-18T13:36:19ZGitHub Agentic autofix：正式提供、警告の修正と検証を統合https://devotd.net/ko/news/github-agentic-autofix-code-scanning-ga-assign-to-copilot/devotd.ko2026-10-18T13:36:19ZGitHub Agentic autofix: 정식 제공, 경고 수정·검증 통합https://devotd.net/zh-hans/news/github-agentic-autofix-code-scanning-ga-assign-to-copilot/devotd.zh-cn2026-10-18T13:36:19ZGitHub Agentic autofix：正式推出，整合警报修复与验证https://devotd.net/en/news/cloudflare-void-open-source-vite-full-stack-platform/devotd.en2026-10-18T13:24:34ZVoid: Vite-based full-stack development, shared team deployment environmenthttps://devotd.net/ja/news/cloudflare-void-open-source-vite-full-stack-platform/devotd.ja2026-10-18T13:24:34ZVoid：Viteベースのフルスタック開発、チーム共有のデプロイ環境を公開https://devotd.net/ko/news/cloudflare-void-open-source-vite-full-stack-platform/devotd.ko2026-10-18T13:24:34ZVoid: Vite 기반 풀스택 개발, 팀 공유 배포 환경 공개https://devotd.net/zh-hans/news/cloudflare-void-open-source-vite-full-stack-platform/devotd.zh-cn2026-10-18T13:24:34ZVoid：基于 Vite 的全栈开发，共享部署环境开源https://devotd.net/en/news/github-copilot-model-deprecation-2026-11-16-opus-gpt-5-6-terra/devotd.en2026-10-18T11:42:49ZGitHub Copilot: Three models retired on November 16, annual subscription exceptionhttps://devotd.net/ja/news/github-copilot-model-deprecation-2026-11-16-opus-gpt-5-6-terra/devotd.ja2026-10-18T11:42:49ZGitHub Copilot：11月16日に3モデルのサポート終了、年間契約の例外https://devotd.net/ko/news/github-copilot-model-deprecation-2026-11-16-opus-gpt-5-6-terra/devotd.ko2026-10-18T11:42:49ZGitHub Copilot: 11월 16일 세 모델 지원 중단, 연간 구독 예외https://devotd.net/zh-hans/news/github-copilot-model-deprecation-2026-11-16-opus-gpt-5-6-terra/devotd.zh-cn2026-10-18T11:42:49ZGitHub Copilot：11月16日停止支持三款模型，年度订阅例外https://devotd.net/en/news/cloudflare-gateway-programmable-workers-dlp/devotd.en2026-10-18T11:25:52ZCloudflare Gateway: User-coded traffic processing, data leak detection validationhttps://devotd.net/ja/news/cloudflare-gateway-programmable-workers-dlp/devotd.ja2026-10-18T11:25:52ZCloudflare Gateway：ユーザーコードによるトラフィック加工、情報漏えい検知の検証https://devotd.net/ko/news/cloudflare-gateway-programmable-workers-dlp/devotd.ko2026-10-18T11:25:52ZCloudflare Gateway: 사용자 코드 기반 트래픽 가공, 정보 유출 탐지 검증https://devotd.net/zh-hans/news/cloudflare-gateway-programmable-workers-dlp/devotd.zh-cn2026-10-18T11:25:52ZCloudflare Gateway：用户代码驱动的流量处理、信息泄露检测验证https://devotd.net/en/news/litellm-1-106-stdio-mcp-disabled-by-default/devotd.en2026-10-18T11:24:16ZLiteLLM v1.106.0: stdio MCP servers disabled by default, one-command Docker startuphttps://devotd.net/ja/news/litellm-1-106-stdio-mcp-disabled-by-default/devotd.ja2026-10-18T11:24:16ZLiteLLM v1.106.0：stdio MCPサーバーをデフォルトで無効化、Dockerを1コマンドで起動https://devotd.net/ko/news/litellm-1-106-stdio-mcp-disabled-by-default/devotd.ko2026-10-18T11:24:16ZLiteLLM v1.106.0: stdio MCP 서버 기본 비활성화, Docker 한 명령 시작https://devotd.net/zh-hans/news/litellm-1-106-stdio-mcp-disabled-by-default/devotd.zh-cn2026-10-18T11:24:16ZLiteLLM v1.106.0：默认禁用 stdio MCP 服务器，一条 Docker 命令启动https://devotd.net/en/news/nvidia-replayssm-hybrid-model-ssm-decoding-state-buffer/devotd.en2026-10-18T09:50:36ZReplaySSM: Separate checkpoints and input buffers, optimize verification and low-precision stateshttps://devotd.net/ja/news/nvidia-replayssm-hybrid-model-ssm-decoding-state-buffer/devotd.ja2026-10-18T09:50:36ZReplaySSM：チェックポイントと入力バッファの分離、検証と低精度状態の最適化https://devotd.net/ko/news/nvidia-replayssm-hybrid-model-ssm-decoding-state-buffer/devotd.ko2026-10-18T09:50:36ZReplaySSM: 체크포인트·입력 버퍼 분리, 검증·저정밀도 상태 최적화https://devotd.net/zh-hans/news/nvidia-replayssm-hybrid-model-ssm-decoding-state-buffer/devotd.zh-cn2026-10-18T09:50:36ZReplaySSM：分离检查点与输入缓冲区，优化验证与低精度状态https://devotd.net/en/news/goose-1-55-ai-gateway-trace-context-auto-compaction/devotd.en2026-10-18T09:40:09Zgoose 1.55.0: Tool call tracing, subagent controlhttps://devotd.net/ja/news/goose-1-55-ai-gateway-trace-context-auto-compaction/devotd.ja2026-10-18T09:40:09Zgoose 1.55.0：ツール呼び出しの追跡、サブエージェントの制御https://devotd.net/ko/news/goose-1-55-ai-gateway-trace-context-auto-compaction/devotd.ko2026-10-18T09:40:09Zgoose 1.55.0: 도구 호출 추적, 하위 에이전트 통제https://devotd.net/zh-hans/news/goose-1-55-ai-gateway-trace-context-auto-compaction/devotd.zh-cn2026-10-18T09:40:09Zgoose 1.55.0：工具调用追踪，子智能体管控https://devotd.net/en/news/github-ai-scan-pull-requests-ga-beyond-codeql/devotd.en2026-10-18T09:31:45ZGitHub AI Scan: Broader security analysis for change reviews, advisory results and credit billinghttps://devotd.net/ja/news/github-ai-scan-pull-requests-ga-beyond-codeql/devotd.ja2026-10-18T09:31:45ZGitHub AI Scan：変更レビューのセキュリティ分析拡大、参考用の結果とクレジット課金https://devotd.net/ko/news/github-ai-scan-pull-requests-ga-beyond-codeql/devotd.ko2026-10-18T09:31:45ZGitHub AI Scan: 변경 검토의 보안 분석 확대, 참고용 결과와 크레딧 과금https://devotd.net/zh-hans/news/github-ai-scan-pull-requests-ga-beyond-codeql/devotd.zh-cn2026-10-18T09:31:45ZGitHub AI Scan：扩大变更审查的安全分析范围，参考性结果与额度计费https://devotd.net/en/news/github-open-source-license-compliance-ga-ruleset-policy-api/devotd.en2026-10-18T08:12:05ZGitHub Enterprise Cloud: Enterprise-wide license policies, compliance checks for coding agentshttps://devotd.net/ja/news/github-open-source-license-compliance-ga-ruleset-policy-api/devotd.ja2026-10-18T08:12:05ZGitHub Enterprise Cloud：企業共通のライセンスポリシー、コーディングエージェント向け準拠チェックhttps://devotd.net/ko/news/github-open-source-license-compliance-ga-ruleset-policy-api/devotd.ko2026-10-18T08:12:05ZGitHub Enterprise Cloud: 기업 공통 라이선스 정책, 코딩 에이전트용 준수 검사https://devotd.net/zh-hans/news/github-open-source-license-compliance-ga-ruleset-policy-api/devotd.zh-cn2026-10-18T08:12:05ZGitHub Enterprise Cloud：企业统一许可证政策、编码智能体合规检查https://devotd.net/en/news/vllm-0-32-0-nvfp4-kv-cache-systemone-breaking-changes/devotd.en2026-10-18T08:04:11ZvLLM v0.32.0: Expanded low-precision caches and universal parsing, tighter security and compatibility changeshttps://devotd.net/ja/news/vllm-0-32-0-nvfp4-kv-cache-systemone-breaking-changes/devotd.ja2026-10-18T08:04:11ZvLLM v0.32.0：低精度キャッシュ・汎用パーサーの拡充、セキュリティ強化と互換性の変更https://devotd.net/ko/news/vllm-0-32-0-nvfp4-kv-cache-systemone-breaking-changes/devotd.ko2026-10-18T08:04:11ZvLLM v0.32.0: 저정밀 캐시·범용 파서 확대, 보안 강화와 호환성 변경https://devotd.net/zh-hans/news/vllm-0-32-0-nvfp4-kv-cache-systemone-breaking-changes/devotd.zh-cn2026-10-18T08:04:11ZvLLM v0.32.0：扩展低精度缓存与通用解析器支持，强化安全并调整兼容性https://devotd.net/en/news/nvidia-run-ai-open-source-kai-scheduler-gpu-fractioning/devotd.en2026-10-18T07:40:15ZNVIDIA Run:ai: Full platform open source, selective component adoptionhttps://devotd.net/ja/news/nvidia-run-ai-open-source-kai-scheduler-gpu-fractioning/devotd.ja2026-10-18T07:40:15ZNVIDIA Run:ai：プラットフォーム全体のオープンソース化、構成要素ごとの選択導入https://devotd.net/ko/news/nvidia-run-ai-open-source-kai-scheduler-gpu-fractioning/devotd.ko2026-10-18T07:40:15ZNVIDIA Run:ai: 전체 플랫폼 오픈소스화, 구성 요소별 선택 도입https://devotd.net/zh-hans/news/nvidia-run-ai-open-source-kai-scheduler-gpu-fractioning/devotd.zh-cn2026-10-18T07:40:15ZNVIDIA Run:ai：整个平台开源，按组件选择引入https://devotd.net/en/news/xolotl-deepseek-v4-1-flash-552b-single-rtx-4090-expert-tiering/devotd.en2026-10-18T03:49:05ZXolotl: Project renamed, large-model inference engines unifiedhttps://devotd.net/ja/news/xolotl-deepseek-v4-1-flash-552b-single-rtx-4090-expert-tiering/devotd.ja2026-10-18T03:49:05ZXolotl：プロジェクト名変更、大型モデル実行エンジン統合https://devotd.net/ko/news/xolotl-deepseek-v4-1-flash-552b-single-rtx-4090-expert-tiering/devotd.ko2026-10-18T03:49:05ZXolotl: 프로젝트 이름 변경, 대형 모델 실행 엔진 통합https://devotd.net/zh-hans/news/xolotl-deepseek-v4-1-flash-552b-single-rtx-4090-expert-tiering/devotd.zh-cn2026-10-18T03:49:05ZXolotl：项目更名，整合大型模型运行引擎https://devotd.net/en/news/llama-cpp-b11717-hexagon-npu-q1-0-q2-0-bonsai-2/devotd.en2026-10-18T03:42:10Zllama.cpp b11716/b11717: Hexagon 1-bit and 2-bit support, row extraction and Hadamard operationshttps://devotd.net/ja/news/llama-cpp-b11717-hexagon-npu-q1-0-q2-0-bonsai-2/devotd.ja2026-10-18T03:42:10Zllama.cpp b11716/b11717：Hexagonの1ビット・2ビット対応、行抽出・アダマール演算の追加https://devotd.net/ko/news/llama-cpp-b11717-hexagon-npu-q1-0-q2-0-bonsai-2/devotd.ko2026-10-18T03:42:10Zllama.cpp b11716/b11717: Hexagon 1비트·2비트 지원, 행 추출·아다마르 연산 추가https://devotd.net/zh-hans/news/llama-cpp-b11717-hexagon-npu-q1-0-q2-0-bonsai-2/devotd.zh-cn2026-10-18T03:42:10Zllama.cpp b11716/b11717：Hexagon 支持 1 位与 2 位权重，新增行提取与哈达玛运算https://devotd.net/en/news/pplx-decider-v1-2-joint-request-decision-model/devotd.en2026-10-18T01:53:31ZPerplexity pplx-decider-v1.2: Grouped questions, restructured training datahttps://devotd.net/ja/news/pplx-decider-v1-2-joint-request-decision-model/devotd.ja2026-10-18T01:53:31ZPerplexity pplx-decider-v1.2：質問の一括処理、学習データの再構成https://devotd.net/ko/news/pplx-decider-v1-2-joint-request-decision-model/devotd.ko2026-10-18T01:53:31ZPerplexity pplx-decider-v1.2: 질문 묶음 처리, 학습 데이터 재구성https://devotd.net/zh-hans/news/pplx-decider-v1-2-joint-request-decision-model/devotd.zh-cn2026-10-18T01:53:31ZPerplexity pplx-decider-v1.2：问题分组处理，训练数据重构https://devotd.net/en/news/moss-image-5b-layout-control-text-rendering-open-weights/devotd.en2026-10-18T01:26:09ZOpenMOSS MOSS-Image-Base·MOSS-Image-SFT: Text and object layout control, base and fine-tuned models releasedhttps://devotd.net/ja/news/moss-image-5b-layout-control-text-rendering-open-weights/devotd.ja2026-10-18T01:26:09ZOpenMOSS MOSS-Image-Base・MOSS-Image-SFT：文字・物体の配置制御、基本・追加学習モデルを公開https://devotd.net/ko/news/moss-image-5b-layout-control-text-rendering-open-weights/devotd.ko2026-10-18T01:26:09ZOpenMOSS MOSS-Image-Base·MOSS-Image-SFT: 글자·사물 배치 제어, 기본·추가 학습 모델 공개https://devotd.net/zh-hans/news/moss-image-5b-layout-control-text-rendering-open-weights/devotd.zh-cn2026-10-18T01:26:09ZOpenMOSS MOSS-Image-Base·MOSS-Image-SFT：文字与物体布局控制，基础与追加训练模型发布https://devotd.net/en/news/cloudflare-openai-crawler-discovery-pilot-results/devotd.en2026-10-18T01:24:25ZCloudflare and OpenAI: Broader page discovery, more efficient search crawling experimentshttps://devotd.net/ja/news/cloudflare-openai-crawler-discovery-pilot-results/devotd.ja2026-10-18T01:24:25ZCloudflare and OpenAI：ページ発見の拡大、検索クロール効率化の実験https://devotd.net/ko/news/cloudflare-openai-crawler-discovery-pilot-results/devotd.ko2026-10-18T01:24:25ZCloudflare and OpenAI: 페이지 발견 확대, 검색 크롤링 효율화 실험https://devotd.net/zh-hans/news/cloudflare-openai-crawler-discovery-pilot-results/devotd.zh-cn2026-10-18T01:24:25ZCloudflare and OpenAI：扩大页面发现，探索提高搜索抓取效率https://devotd.net/en/news/dspine-qwen3-8b-parallel-drafter-speculative-decoding/devotd.en2026-10-17T23:21:48ZDSpine Qwen3-8B: Parallel draft model released, inference code providedhttps://devotd.net/ja/news/dspine-qwen3-8b-parallel-drafter-speculative-decoding/devotd.ja2026-10-17T23:21:48ZDSpine Qwen3-8B：並列ドラフトモデル公開、推論コード提供https://devotd.net/ko/news/dspine-qwen3-8b-parallel-drafter-speculative-decoding/devotd.ko2026-10-17T23:21:48ZDSpine Qwen3-8B: 병렬 초안 모델 공개, 추론 코드 제공https://devotd.net/zh-hans/news/dspine-qwen3-8b-parallel-drafter-speculative-decoding/devotd.zh-cn2026-10-17T23:21:48ZDSpine Qwen3-8B：公开并行草稿模型，提供推理代码https://devotd.net/en/news/nvidia-indic-pretraining-2-8t-tokens-open-pipeline/devotd.en2026-10-17T23:18:14ZNemotron-pretraining-Indic-v1: 2.8 trillion-token corpus, open data generation pipelinehttps://devotd.net/ja/news/nvidia-indic-pretraining-2-8t-tokens-open-pipeline/devotd.ja2026-10-17T23:18:14ZNemotron-pretraining-Indic-v1：2.8兆トークンのコーパス、公開データ生成パイプラインhttps://devotd.net/ko/news/nvidia-indic-pretraining-2-8t-tokens-open-pipeline/devotd.ko2026-10-17T23:18:14ZNemotron-pretraining-Indic-v1: 2.8조 토큰 코퍼스, 공개 데이터 생성 파이프라인https://devotd.net/zh-hans/news/nvidia-indic-pretraining-2-8t-tokens-open-pipeline/devotd.zh-cn2026-10-17T23:18:14ZNemotron-pretraining-Indic-v1：2.8万亿 token 语料库，公开数据生成流程https://devotd.net/en/news/cloudflare-verified-bots-continuous-accountability/devotd.en2026-10-17T23:17:47ZCloudflare Verified Bots: Continuous Behavior Checks, Customer-Defined Trust Criteriahttps://devotd.net/ja/news/cloudflare-verified-bots-continuous-accountability/devotd.ja2026-10-17T23:17:47ZCloudflareの認証済みボット：継続的な行動検証、顧客ごとの信頼基準https://devotd.net/ko/news/cloudflare-verified-bots-continuous-accountability/devotd.ko2026-10-17T23:17:47ZCloudflare 인증 봇: 지속적 행동 검증, 고객별 신뢰 기준https://devotd.net/zh-hans/news/cloudflare-verified-bots-continuous-accountability/devotd.zh-cn2026-10-17T23:17:47ZCloudflare 已验证机器人：持续行为验证，客户自定信任标准https://devotd.net/en/news/llama-cpp-b11714-discover-models-hf-hub-draft-sidecars/devotd.en2026-10-17T21:33:33Zllama.cpp b11714: Web model discovery and auxiliary file selection, download pause and resumehttps://devotd.net/ja/news/llama-cpp-b11714-discover-models-hf-hub-draft-sidecars/devotd.ja2026-10-17T21:33:33Zllama.cpp b11714：Webでのモデル探索と補助ファイル選択、ダウンロードの一時停止と再開https://devotd.net/ko/news/llama-cpp-b11714-discover-models-hf-hub-draft-sidecars/devotd.ko2026-10-17T21:33:33Zllama.cpp b11714: 웹 모델 탐색과 보조 파일 선택, 다운로드 일시정지와 재개https://devotd.net/zh-hans/news/llama-cpp-b11714-discover-models-hf-hub-draft-sidecars/devotd.zh-cn2026-10-17T21:33:33Zllama.cpp b11714：网页模型探索与辅助文件选择，下载暂停与恢复https://devotd.net/en/news/nemotron-labs-dory-1-0-4b-looped-elastic-compute/devotd.en2026-10-17T21:32:19ZNVIDIA-Nemotron-Labs-Dory1.0-4B: Adjustable loop counts, last-loop cachinghttps://devotd.net/ja/news/nemotron-labs-dory-1-0-4b-looped-elastic-compute/devotd.ja2026-10-17T21:32:19ZNVIDIA-Nemotron-Labs-Dory1.0-4B：反復回数の調整、最終反復のキャッシュhttps://devotd.net/ko/news/nemotron-labs-dory-1-0-4b-looped-elastic-compute/devotd.ko2026-10-17T21:32:19ZNVIDIA-Nemotron-Labs-Dory1.0-4B: 반복 횟수 조절, 마지막 반복 캐시https://devotd.net/zh-hans/news/nemotron-labs-dory-1-0-4b-looped-elastic-compute/devotd.zh-cn2026-10-17T21:32:19ZNVIDIA-Nemotron-Labs-Dory1.0-4B：调整循环次数，缓存最后一轮https://devotd.net/en/news/opencode-console-provider-key-breach-distillation-attack/devotd.en2026-10-17T21:17:10ZOpenCode Console: External key exposure, provider key revocation and replacementhttps://devotd.net/ja/news/opencode-console-provider-key-breach-distillation-attack/devotd.ja2026-10-17T21:17:10ZOpenCode Console：外部キーの露出、プロバイダーキーの失効・交換https://devotd.net/ko/news/opencode-console-provider-key-breach-distillation-attack/devotd.ko2026-10-17T21:17:10ZOpenCode Console: 외부 키 노출, 제공자 키 폐기·교체https://devotd.net/zh-hans/news/opencode-console-provider-key-breach-distillation-attack/devotd.zh-cn2026-10-17T21:17:10ZOpenCode Console：外部密钥暴露，服务商密钥撤销与更换https://devotd.net/en/news/nemotron-3-5-bolt-74b-a9b-24gb-2bit-local-agent/devotd.en2026-10-17T17:48:03ZNVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB: Compression through 2-bit training, local vision-language agentshttps://devotd.net/ja/news/nemotron-3-5-bolt-74b-a9b-24gb-2bit-local-agent/devotd.ja2026-10-17T17:48:03ZNVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB：2ビット学習による圧縮、ローカル視覚言語エージェントhttps://devotd.net/ko/news/nemotron-3-5-bolt-74b-a9b-24gb-2bit-local-agent/devotd.ko2026-10-17T17:48:03ZNVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB: 2비트 학습 기반 압축, 로컬 시각언어 에이전트https://devotd.net/zh-hans/news/nemotron-3-5-bolt-74b-a9b-24gb-2bit-local-agent/devotd.zh-cn2026-10-17T17:48:03ZNVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB：基于2位训练的压缩，本地视觉语言智能体https://devotd.net/en/news/reflection-beam-1-0-open-weights-self-hosting/devotd.en2026-10-17T17:29:39ZBeam 1.0: Weights in three precisions, Apache 2.0 self-hostinghttps://devotd.net/ja/news/reflection-beam-1-0-open-weights-self-hosting/devotd.ja2026-10-17T17:29:39ZBeam 1.0：3つの精度の重み公開、Apache 2.0によるセルフホスティングhttps://devotd.net/ko/news/reflection-beam-1-0-open-weights-self-hosting/devotd.ko2026-10-17T17:29:39ZBeam 1.0: 세 정밀도 가중치 공개, Apache 2.0 기반 자체 호스팅https://devotd.net/zh-hans/news/reflection-beam-1-0-open-weights-self-hosting/devotd.zh-cn2026-10-17T17:29:39ZBeam 1.0：三种精度权重公开，基于 Apache 2.0 的自托管https://devotd.net/en/news/claude-code-2-1-302-auto-mode-classifier-idle-compaction/devotd.en2026-10-17T17:21:54ZClaude Code 2.1.302: Auto mode and idle compaction settings, wait and image read limitshttps://devotd.net/ja/news/claude-code-2-1-302-auto-mode-classifier-idle-compaction/devotd.ja2026-10-17T17:21:54ZClaude Code 2.1.302：自動モード・アイドル時圧縮の設定、待機・画像読み取りの上限https://devotd.net/ko/news/claude-code-2-1-302-auto-mode-classifier-idle-compaction/devotd.ko2026-10-17T17:21:54ZClaude Code 2.1.302: 자동 모드·유휴 압축 설정, 대기·이미지 읽기 상한https://devotd.net/zh-hans/news/claude-code-2-1-302-auto-mode-classifier-idle-compaction/devotd.zh-cn2026-10-17T17:21:54ZClaude Code 2.1.302：自动模式与空闲压缩设置，等待与图像读取上限https://devotd.net/en/news/ktransformers-0-7-2-deepseek-v4-flash-mxfp4-lora/devotd.en2026-10-17T15:19:31ZKTransformers v0.7.2: LoRA training for compressed CPU experts, training tools linked to inference adaptershttps://devotd.net/ja/news/ktransformers-0-7-2-deepseek-v4-flash-mxfp4-lora/devotd.ja2026-10-17T15:19:31ZKTransformers v0.7.2：CPU圧縮エキスパートのLoRA学習、学習ツールと推論アダプターの連携https://devotd.net/ko/news/ktransformers-0-7-2-deepseek-v4-flash-mxfp4-lora/devotd.ko2026-10-17T15:19:31ZKTransformers v0.7.2: CPU 압축 전문가 LoRA 학습, 학습 도구와 추론 어댑터 연결https://devotd.net/zh-hans/news/ktransformers-0-7-2-deepseek-v4-flash-mxfp4-lora/devotd.zh-cn2026-10-17T15:19:31ZKTransformers v0.7.2：CPU 压缩专家 LoRA 训练，训练工具与推理适配器衔接https://devotd.net/en/news/intel-llm-scaler-vllm-1-5-qwen-gemma-4-fp8-mtp/devotd.en2026-10-17T15:18:04ZIntel LLM-Scaler vLLM 1.5: Updated inference engine, broader model and quantization supporthttps://devotd.net/ja/news/intel-llm-scaler-vllm-1-5-qwen-gemma-4-fp8-mtp/devotd.ja2026-10-17T15:18:04ZIntel LLM-Scaler vLLM 1.5：推論エンジンの更新、モデル・量子化対応の拡大https://devotd.net/ko/news/intel-llm-scaler-vllm-1-5-qwen-gemma-4-fp8-mtp/devotd.ko2026-10-17T15:18:04ZIntel LLM-Scaler vLLM 1.5: 추론 엔진 기반 갱신, 모델·양자화 지원 확대https://devotd.net/zh-hans/news/intel-llm-scaler-vllm-1-5-qwen-gemma-4-fp8-mtp/devotd.zh-cn2026-10-17T15:18:04ZIntel LLM-Scaler vLLM 1.5：更新底层推理引擎，扩展模型与量化支持https://devotd.net/en/news/granite-switch-4-1-decision-guardian-speech-one-endpoint/devotd.en2026-10-17T13:27:20ZGranite Switch 4.1: Unified chat, decision-making and speech recognition, open weightshttps://devotd.net/ja/news/granite-switch-4-1-decision-guardian-speech-one-endpoint/devotd.ja2026-10-17T13:27:20ZGranite Switch 4.1：対話・判断・音声認識の統合、公開ウェイトhttps://devotd.net/ko/news/granite-switch-4-1-decision-guardian-speech-one-endpoint/devotd.ko2026-10-17T13:27:20ZGranite Switch 4.1: 대화·판단·음성 인식 통합, 공개 가중치https://devotd.net/zh-hans/news/granite-switch-4-1-decision-guardian-speech-one-endpoint/devotd.zh-cn2026-10-17T13:27:20ZGranite Switch 4.1：集成对话、决策与语音识别，开放权重https://devotd.net/en/news/llama-cpp-b11709-kolibri-1-gguf-support/devotd.en2026-10-17T13:20:41Zllama.cpp b11709: Kolibri-1 support, GGUF conversion path addedhttps://devotd.net/ja/news/llama-cpp-b11709-kolibri-1-gguf-support/devotd.ja2026-10-17T13:20:41Zllama.cpp b11709：Kolibri-1対応、GGUF変換経路の追加https://devotd.net/ko/news/llama-cpp-b11709-kolibri-1-gguf-support/devotd.ko2026-10-17T13:20:41Zllama.cpp b11709: Kolibri-1 지원, GGUF 변환 경로 추가https://devotd.net/zh-hans/news/llama-cpp-b11709-kolibri-1-gguf-support/devotd.zh-cn2026-10-17T13:20:41Zllama.cpp b11709：支持 Kolibri-1，新增 GGUF 转换路径https://devotd.net/en/news/strata-0-1-45-qwen3-6-vision-parallel-requests/devotd.en2026-10-17T12:24:07ZStrata 0.1.45: Qwen3.6-35B-A3B image input, up to 8 concurrent requests and faster generationhttps://devotd.net/ja/news/strata-0-1-45-qwen3-6-vision-parallel-requests/devotd.ja2026-10-17T12:24:07ZStrata 0.1.45：Qwen3.6-35B-A3Bの画像入力、最大8件の同時リクエストと生成の高速化https://devotd.net/ko/news/strata-0-1-45-qwen3-6-vision-parallel-requests/devotd.ko2026-10-17T12:24:07ZStrata 0.1.45: Qwen3.6-35B-A3B 이미지 입력, 최대 8개 동시 요청과 생성 가속https://devotd.net/zh-hans/news/strata-0-1-45-qwen3-6-vision-parallel-requests/devotd.zh-cn2026-10-17T12:24:07ZStrata 0.1.45：Qwen3.6-35B-A3B 图像输入，最多 8 个并发请求与生成加速https://devotd.net/en/news/fish-audio-drama-3-35b-a3b-open-tts/devotd.en2026-10-17T11:27:01ZFish Audio Drama 3: Multi-speaker direction control, Segment-level audio editinghttps://devotd.net/ja/news/fish-audio-drama-3-35b-a3b-open-tts/devotd.ja2026-10-17T11:27:01ZFish Audio Drama 3：複数話者の演出制御、区間単位の音声編集https://devotd.net/ko/news/fish-audio-drama-3-35b-a3b-open-tts/devotd.ko2026-10-17T11:27:01ZFish Audio Drama 3: 다화자 연출 제어, 구간 단위 음성 편집https://devotd.net/zh-hans/news/fish-audio-drama-3-35b-a3b-open-tts/devotd.zh-cn2026-10-17T11:27:01ZFish Audio Drama 3：多说话人演绎控制，分段语音编辑https://devotd.net/en/news/flux-3-image-klein-5-6b-open-weights/devotd.en2026-10-17T11:23:25ZFLUX 3 Image Klein: Open weights, 5.6 billion-parameter generatorhttps://devotd.net/ja/news/flux-3-image-klein-5-6b-open-weights/devotd.ja2026-10-17T11:23:25ZFLUX 3 Image Klein：重み公開、56億パラメータの生成器https://devotd.net/ko/news/flux-3-image-klein-5-6b-open-weights/devotd.ko2026-10-17T11:23:25ZFLUX 3 Image Klein: 가중치 공개, 56억 파라미터 생성기https://devotd.net/zh-hans/news/flux-3-image-klein-5-6b-open-weights/devotd.zh-cn2026-10-17T11:23:25ZFLUX 3 Image Klein：公开权重，56亿参数生成器https://devotd.net/en/news/deepseek-v4-1-flash-1017-open-weights-search-fetch/devotd.en2026-10-17T11:23:23ZDeepSeek-V4.1-Flash-1017: Improved coding performance and speed, public Search & Fetch APIshttps://devotd.net/ja/news/deepseek-v4-1-flash-1017-open-weights-search-fetch/devotd.ja2026-10-17T11:23:23ZDeepSeek-V4.1-Flash-1017：コーディング性能・速度の改善、検索・取得APIの公開https://devotd.net/ko/news/deepseek-v4-1-flash-1017-open-weights-search-fetch/devotd.ko2026-10-17T11:23:23ZDeepSeek-V4.1-Flash-1017: 코딩 성능·속도 개선, 검색·가져오기 API 공개https://devotd.net/zh-hans/news/deepseek-v4-1-flash-1017-open-weights-search-fetch/devotd.zh-cn2026-10-17T11:23:23ZDeepSeek-V4.1-Flash-1017：提升编程性能与速度，开放搜索与抓取 APIhttps://devotd.net/en/news/openai-agents-python-0-24-mcp-annotations-tool-errors/devotd.en2026-10-17T09:28:13ZOpenAI Agents SDK for Python 0.24.0: MCP tool annotation exposure, run-wide error formattinghttps://devotd.net/ja/news/openai-agents-python-0-24-mcp-annotations-tool-errors/devotd.ja2026-10-17T09:28:13ZOpenAI Agents SDK for Python 0.24.0：MCPツールの説明情報の公開、実行全体でのエラー整形https://devotd.net/ko/news/openai-agents-python-0-24-mcp-annotations-tool-errors/devotd.ko2026-10-17T09:28:13ZOpenAI Agents SDK for Python 0.24.0: MCP 도구 설명 노출, 실행 전체 오류 서식화https://devotd.net/zh-hans/news/openai-agents-python-0-24-mcp-annotations-tool-errors/devotd.zh-cn2026-10-17T09:28:13ZOpenAI Agents SDK for Python 0.24.0：公开 MCP 工具注释，统一整个运行的错误格式https://devotd.net/en/news/lfm2-5-24b-a2b-on-device-agent-moe/devotd.en2026-10-17T08:47:15ZLFM2.5-24B-A2B: On-device reasoning and tool use, reduced computation through partial activationhttps://devotd.net/ja/news/lfm2-5-24b-a2b-on-device-agent-moe/devotd.ja2026-10-17T08:47:15ZLFM2.5-24B-A2B：デバイス内の推論・ツール活用、一部の活性化で計算削減https://devotd.net/ko/news/lfm2-5-24b-a2b-on-device-agent-moe/devotd.ko2026-10-17T08:47:15ZLFM2.5-24B-A2B: 기기 내 추론·도구 활용, 부분 활성화로 계산 절감https://devotd.net/zh-hans/news/lfm2-5-24b-a2b-on-device-agent-moe/devotd.zh-cn2026-10-17T08:47:15ZLFM2.5-24B-A2B：设备端推理与工具使用，部分激活以减少计算https://devotd.net/en/news/minicpm-v-4-7-35b-a3b-on-device-moe-vision/devotd.en2026-10-17T08:45:38ZMiniCPM-V 4.7 35B-A3B: The series' first MoE, on-device expert prefetchinghttps://devotd.net/ja/news/minicpm-v-4-7-35b-a3b-on-device-moe-vision/devotd.ja2026-10-17T08:45:38ZMiniCPM-V 4.7 35B-A3B：シリーズ初のMoE、端末上でのエキスパート先読みhttps://devotd.net/ko/news/minicpm-v-4-7-35b-a3b-on-device-moe-vision/devotd.ko2026-10-17T08:45:38ZMiniCPM-V 4.7 35B-A3B: 시리즈 첫 MoE, 기기 내 전문가 미리 읽기https://devotd.net/zh-hans/news/minicpm-v-4-7-35b-a3b-on-device-moe-vision/devotd.zh-cn2026-10-17T08:45:38ZMiniCPM-V 4.7 35B-A3B：系列首款 MoE，端侧专家预读取https://devotd.net/en/news/offrail-27b-uncensored-bonsai-2-mtp-graft/devotd.en2026-10-17T05:50:21ZOffrail-27B-Uncensored: Uncensored ternary model integration, multi-token prediction head transplanthttps://devotd.net/ja/news/offrail-27b-uncensored-bonsai-2-mtp-graft/devotd.ja2026-10-17T05:50:21ZOffrail-27B-Uncensored：非検閲3値モデルの統合、多トークン予測ヘッドの移植https://devotd.net/ko/news/offrail-27b-uncensored-bonsai-2-mtp-graft/devotd.ko2026-10-17T05:50:21ZOffrail-27B-Uncensored: 비검열 삼진 모델 통합, 다중 토큰 예측 헤드 이식https://devotd.net/zh-hans/news/offrail-27b-uncensored-bonsai-2-mtp-graft/devotd.zh-cn2026-10-17T05:50:21ZOffrail-27B-Uncensored：集成无审查三值模型，移植多令牌预测头https://devotd.net/en/news/unsloth-0-1-908-beta-nemotron-muse-spark-prompt-cache/devotd.en2026-10-17T05:16:48ZUnsloth v0.1.908-beta: Local Nemotron 3.5 and Muse Spark Open, improved prompt caching for coding agentshttps://devotd.net/ja/news/unsloth-0-1-908-beta-nemotron-muse-spark-prompt-cache/devotd.ja2026-10-17T05:16:48ZUnsloth v0.1.908-beta：Nemotron 3.5・Muse Spark Openのローカル実行、コーディングエージェントのプロンプトキャッシュ改善https://devotd.net/ko/news/unsloth-0-1-908-beta-nemotron-muse-spark-prompt-cache/devotd.ko2026-10-17T05:16:48ZUnsloth v0.1.908-beta: Nemotron 3.5·Muse Spark Open 로컬 실행, 코딩 에이전트 프롬프트 캐싱 개선https://devotd.net/zh-hans/news/unsloth-0-1-908-beta-nemotron-muse-spark-prompt-cache/devotd.zh-cn2026-10-17T05:16:48ZUnsloth v0.1.908-beta：本地运行 Nemotron 3.5 与 Muse Spark Open，改进编码智能体提示缓存https://devotd.net/en/news/ray-2-59-llm-ga-token-auth-default/devotd.en2026-10-17T04:38:52ZRay 2.59: Language model data processing and serving reach general availability, token authentication enabled by default for local clustershttps://devotd.net/ja/news/ray-2-59-llm-ga-token-auth-default/devotd.ja2026-10-17T04:38:52ZRay 2.59：言語モデルのデータ処理・サービングが正式リリース、ローカルクラスタのトークン認証をデフォルト化https://devotd.net/ko/news/ray-2-59-llm-ga-token-auth-default/devotd.ko2026-10-17T04:38:52ZRay 2.59: 언어 모델 데이터 처리·서빙 정식 출시, 로컬 클러스터 토큰 인증 기본화https://devotd.net/zh-hans/news/ray-2-59-llm-ga-token-auth-default/devotd.zh-cn2026-10-17T04:38:52ZRay 2.59：语言模型数据处理与服务正式发布，默认启用本地集群令牌认证https://devotd.net/en/news/claude-code-2-1-300-mods-otel-advisor/devotd.en2026-10-17T04:27:21ZClaude Code 2.1.300: Separate prompt decorations, GenAI observability attributeshttps://devotd.net/ja/news/claude-code-2-1-300-mods-otel-advisor/devotd.ja2026-10-17T04:27:21ZClaude Code 2.1.300：入力欄の装飾分離、生成AIの観測属性https://devotd.net/ko/news/claude-code-2-1-300-mods-otel-advisor/devotd.ko2026-10-17T04:27:21ZClaude Code 2.1.300: 입력창 장식 분리, 생성형 인공지능 관측 속성https://devotd.net/zh-hans/news/claude-code-2-1-300-mods-otel-advisor/devotd.zh-cn2026-10-17T04:27:21ZClaude Code 2.1.300：输入框装饰分离，生成式人工智能观测属性https://devotd.net/en/news/sglang-0-5-21-release-new-models/devotd.en2026-10-17T00:53:14ZSGLang v0.5.21: Expanded model support, changed cache and request handling defaultshttps://devotd.net/ja/news/sglang-0-5-21-release-new-models/devotd.ja2026-10-17T00:53:14ZSGLang v0.5.21：新規モデル対応の拡大、キャッシュ・リクエスト処理のデフォルト変更https://devotd.net/ko/news/sglang-0-5-21-release-new-models/devotd.ko2026-10-17T00:53:14ZSGLang v0.5.21: 신규 모델 지원 확대, 캐시·요청 처리 기본값 변경https://devotd.net/zh-hans/news/sglang-0-5-21-release-new-models/devotd.zh-cn2026-10-17T00:53:14ZSGLang v0.5.21：扩展新模型支持，调整缓存与请求处理默认行为https://devotd.net/en/news/open-webui-0-12-3-end-to-end-encrypted-chats/devotd.en2026-10-17T00:50:55ZOpen WebUI 0.12.3: Browser end-to-end encryption, temporary chat permissionshttps://devotd.net/ja/news/open-webui-0-12-3-end-to-end-encrypted-chats/devotd.ja2026-10-17T00:50:55ZOpen WebUI 0.12.3：ブラウザーでのエンドツーエンド暗号化、一時チャットの権限管理https://devotd.net/ko/news/open-webui-0-12-3-end-to-end-encrypted-chats/devotd.ko2026-10-17T00:50:55ZOpen WebUI 0.12.3: 브라우저 종단간 암호화, 임시 대화 권한 관리https://devotd.net/zh-hans/news/open-webui-0-12-3-end-to-end-encrypted-chats/devotd.zh-cn2026-10-17T00:50:55ZOpen WebUI 0.12.3：浏览器端到端加密，临时对话权限管理https://devotd.net/en/news/hy-memory-30b-a3b-open-memory-writer/devotd.en2026-10-16T18:05:56ZHy-Memory-30B-A3B: Memory writing with 3 billion active parameters, integration with existing memory frameworkshttps://devotd.net/ja/news/hy-memory-30b-a3b-open-memory-writer/devotd.ja2026-10-16T18:05:56ZHy-Memory-30B-A3B：アクティブ30億パラメータで記憶作成に特化、既存の記憶フレームワークと連携https://devotd.net/ko/news/hy-memory-30b-a3b-open-memory-writer/devotd.ko2026-10-16T18:05:56ZHy-Memory-30B-A3B: 활성 30억 파라미터의 기억 작성 특화, 기존 기억 프레임워크 연동https://devotd.net/zh-hans/news/hy-memory-30b-a3b-open-memory-writer/devotd.zh-cn2026-10-16T18:05:56ZHy-Memory-30B-A3B：30亿活跃参数专攻记忆写入，接入现有记忆框架