to - 钛刻 - 科技风向旗 - 深度刻画技术趋势,引领数字未来 - 第75页 - 钛刻科技

gpt-5.5 PreToolUse hook (failed) 如何解决？

PreToolUse hook (failed) error: hook exited with code 1 Windows 官方codex cli 1 个帖子 - 1 位参与者阅读完整话题

相关专题

tech linux.do 2026-04-25 00:17:30+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:54:14+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:54:14+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:54:14+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:47:32+08:00

“大 F”键盘终结：FILCO（斐尔可）母公司 Diatech 宣布停业

IT之家 4 月 24 日消息，俗称“大 F”的 FILCO（斐尔可）以其 Majestouch“圣手”系列机械键盘闻名，是键盘行业早期的代表性品牌之一，已有 34 年历史。不过就在本周三，FILCO 母公司 Diatech 宣布停业，这也意味着 “大 F”键盘已正式退出历史舞

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech www.ithome.com 2026-04-24 23:40:48+08:00

从早上九点到现在，用了1.2亿token

今天写了一天代码，腰酸背痛的，一看cpa，今天竟然用了1.2亿token，买的日抛team回本了哈哈哈 14 个帖子 - 13 位参与者阅读完整话题

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech linux.do 2026-04-24 23:39:27+08:00

有用axonhub的佬吗，目前能转发gpt image 2吗？

我配上去后拿cherry studio测试报错：failed to transform request: invalid request: image is not supported 不知道是我配置问题，还是暂时还不支持 1 个帖子 - 1 位参与者阅读完整话题

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech linux.do 2026-04-24 23:33:36+08:00

速登！免费！阿里云百炼上下deepseek4，速度快到飞起！

不是很多但速度是真的快啊！ flash|每秒175 tokens pro|每秒81 tokens 我觉得等大家都完善配置了速度还能起飞 12 个帖子 - 9 位参与者阅读完整话题

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech linux.do 2026-04-24 23:24:54+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:03:46+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 23:03:46+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 22:48:43+08:00

我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech www.v2ex.com 2026-04-24 22:48:43+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 22:41:38+08:00

Meta 与亚马逊 AWS 达成合作，部署数千万 Graviton 核心构建 AI 基础设施

IT之家 4 月 24 日消息，Meta 今天与亚马逊 AWS 达成合作协议，大规模部署 AWS Graviton 处理器，为下一代 AI 打下基础。据介绍，Meta 将首先部署数千万个 Graviton 核心，随后根据 AI 能力增长进度灵活扩展。虽然两家公司暂未披露具体合作

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech www.ithome.com 2026-04-24 22:40:06+08:00

google pay买claude或gpt的订阅，会看卡的类型吗

如题，我如果用bybit的u卡已经绑定了google pay，然后我拿安卓手机在google store下载claude或者gpt进行订阅能够成功吗。另外如果封号了，那么这个费用是会把订阅的金额全额退回吗，如果在用了几天就封号了，那么还全额退了那不就是白嫖了几天嘛，没有这方面的经

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech linux.do 2026-04-24 22:23:10+08:00

API方式登录的Codex Desktop能使用插件功能吗

我发现主分支里的Superpowers是可以用插件的方式安装但是codex Desktop无法在API模式使用 3 个帖子 - 3 位参与者阅读完整话题

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech linux.do 2026-04-24 22:08:17+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 21:41:54+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 21:41:54+08:00

[Local LLM] 我做了个工具让 8GB 显卡跑 30B 模型从 3 tok/s 提到 21 tok/s，记录一下技术发现

最近在折腾本地大模型，发现一个核心问题：Ollama 和 LM Studio 能让模型跑起来，但参数全靠猜——上下文长度、KV cache 类型、MoE expert 放哪、ubatch 多大……用默认参数基本是在浪费显卡。于是做了个工具自动找最优配置，过程中踩了不少坑，记录一

相关专题

Health Client 专题内容 Interface Fitness Experience Coupon 专题内容 Strategy 专题内容 Brand Customization Community Target Sale Alert Automation 专题内容 Saving Supplier Collaborate Kpi Domain Local Reporting Schedu...游戏 Upload Budget Customer Sale 专题内容 Conversion Community Mobile Performance 专题内容 Customer Behavior Objective Network Theme Visitor Music Folde...Digital Planning Security Partner 专题内容影视 Mobile Restaurant Help Budget Client Conversion Management...Software File Analytics 专题内容 Income Forum Performance 专题内容 File Like Planning Partner Unsubscribe Project Button 专题内容影视 Interface Partner 专题内容影视 Theme Advertising Tool Task Food Deal Change Affordable 专题内容 Study Folder Excellence 专题内容 Tracking Navigation Campaign Dashboard Study Recommendation 专题内容 Section Notification Tool Status Optimization Education 专题内容 Marketing 专题内容 Team Funnel Navigation Careers Screen Optimization 专题内容

tech v2ex.com 2026-04-24 21:41:54+08:00

to - 钛刻 - 科技风向旗 - 深度刻画技术趋势,引领数字未来 - 第75页 - 钛刻科技 | TCTI.cn

相关标签