模型 - 钛刻 - 科技风向旗 - 深度刻画技术趋势,引领数字未来 - 第68页 - 钛刻科技

大伙有想过二次训练吗？

比如我下载了一个模型。然后再把我所有文档交给它，二次训练。那么，是不是就没必要 RAG 了。通过这个模型，我就能提问了嘛，毕竟，我的基因已经嵌入进去了。

相关专题

tech www.v2ex.com 2026-04-24 19:58:36+08:00

美团万亿级大模型 LongCat-2.0-Preview 开放测试，全程基于国产算力集群训练

美团万亿级大模型 LongCat-2.0-Preview 开放测试，全程基于国产算力集群训练 - IT之家 6 个帖子 - 6 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:45:09+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

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.v2ex.com 2026-04-24 19:41:32+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

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech v2ex.com 2026-04-24 19:41:32+08:00

有没有稳定好用的中转器，除了cpa sub2api

这几天真是被sub2api cpa气的不行，功能做的实在是太差了先是cpa有模型添加bug，我明明加了5.3codex，模型列表就是不显示又加了个5.4，你猜怎么着，5.3codex又回来了然后是sub2api，格式转换一坨，工具调用全出错，cpa也差不多少，我firewo

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:33:49+08:00

使用claudecode最新版遇到的问题

为什么我用/model这样切换模型会直接卡死，有佬友可以指点一下吗 2 个帖子 - 2 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:32:08+08:00

gpt 5.5 有官方敏感词拦截了？

之前破限的好好的，今天使用5.5以后，经常性弹出这个来。是不是现在模型添加了敏感检测了？ 3 个帖子 - 3 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:24:54+08:00

deepseek网页端是不是v4

deepseek网页端是不是v4模型？肯定不可能是v4-pro，但是v4吗？不会还是v3.2吧有佬知道吗？ 3 个帖子 - 2 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:14:56+08:00

现在安卓哪个客户端支持cpa里的codex的gpt–image-2模型的？

试了下rikkahub不能正常调用，不知道有没有别的客户端可以支持 7 个帖子 - 3 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:07:53+08:00

一样的上下文, Claude占用Token是其他模型的1.5倍

以前在OpenCode中使用Claude和GPT的时候就发现. 同样GPT的聊天180ktoken, 一切到Claude就变成260ktoken, 我一度以为是GPT比较省Token. 现在用CC, 用Claude聊天, 显示160ktoken, 切到GLM5.1变成100kto

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:04:52+08:00

codex windows端频繁报错

RT，今天让5.5模型帮忙生成一些几十个emoji表情，思考速度是中，标准模式和快速模式都试过，但是依旧是生成几个之后就开始报错，只能重新继续任务，下面是报错信息，跟我的网络环境有关系吗？ stream disconnected before completion: An err

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 19:03:59+08:00

大伙有想过二次训练吗？

比如我下载了一个模型。然后再把我所有文档交给它，二次训练。那么，是不是就没必要 RAG 了。通过这个模型，我就能提问了嘛，毕竟，我的基因已经嵌入进去了。

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.v2ex.com 2026-04-24 18:58:36+08:00

大伙有想过二次训练吗？

比如我下载了一个模型。然后再把我所有文档交给它，二次训练。那么，是不是就没必要 RAG 了。通过这个模型，我就能提问了嘛，毕竟，我的基因已经嵌入进去了。

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.v2ex.com 2026-04-24 18:58:36+08:00

求问还有木有邀请送国产大模型token的平台

萌新求问目前还有没有便宜的相对稳定邀请送token，硅基流动用完了新的太贵七牛云被封了一个不敢用了 1 个帖子 - 1 位参与者阅读完整话题

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech linux.do 2026-04-24 18:48:40+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

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech v2ex.com 2026-04-24 18:48:29+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

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech v2ex.com 2026-04-24 18:48:29+08:00

[ 模型对比 ] 用手里有的几个模型做了一个小对比

实现一个比较常规机械时钟，实现的效果如下链接直达

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.v2ex.com 2026-04-24 18:44:09+08:00

[ 模型对比 ] 用手里有的几个模型做了一个小对比

实现一个比较常规机械时钟，实现的效果如下链接直达

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.v2ex.com 2026-04-24 18:44:09+08:00

[程序员] [ 模型对比 ] 用手里有的几个模型做了一个小对比

实现一个比较常规机械时钟，实现的效果如下链接直达

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech v2ex.com 2026-04-24 18:40:15+08:00

美团万亿级大模型 LongCat-2.0-Preview 开放测试，全程基于国产算力集群训练

IT之家 4 月 24 日消息，据《科创板日报》今日报道，美团新一代基础大模型 LongCat-2.0-Preview 已经开放邀请测试，该模型总参数规模突破万亿，量级跻身全球顶尖大模型行列。这款尚未公布正式命名的大模型参数量级达到万亿级别，采用了混合专家（MoE）的先进架构，

相关专题

Guide Customer Rating Economy 专题内容 Form Social Backup 专题内容 Value Settings Folder Integration 专题内容 Investment Achievement Performance Restore Optimization 专题内容 Whitepaper User Reminder 专题内容 Fitness Settings Browser Domain 专题内容 Media 专题内容 Enterprise Automation Demographic Training Accessibility Ente...Client Vacation Download Services Training Education 专题内容 Presentation 专题内容 Quality Domain Media Landing Contact Project 专题内容 Deadline Customer Story Database 专题内容 Management Accessibility 专题内容 Webinar Change Automation 专题内容 Alert Advertising 专题内容 Tactic Community Login Study Creative Tool Dashboard 专题内容 Enterprise Dashboard Deal Landing Cloud Entertainment Trainin...Income Ranking Restaurant Automation Marketing Database Backu...Network Partner Sync Project Productivity Target Presentation...Design Story Expensive Resource Status Engagement Change Devi...

tech www.ithome.com 2026-04-24 18:38:14+08:00

模型 - 钛刻 - 科技风向旗 - 深度刻画技术趋势,引领数字未来 - 第68页 - 钛刻科技 | TCTI.cn

相关标签