注册并分享邀请链接,可获得视频播放与邀请奖励。

与「MiniMax」相关的搜索结果

MiniMax 贴吧
一个关键词就是一个贴吧,路径全站唯一。
创建贴吧
用户
未找到
包含 MiniMax 的内容
继腾讯WorKBuddy之后 豆包工作是国内大厂中 第二个变成模型商店的 办公智能体 不过相比WorKBuddy 里面少了MiniMax 也没有小米阶跃星辰之类的 看来字节对模型筛选的标准很高 只放高水平的第三方旗舰模型 虽然还不支持用户 自定义第三方API 但豆包工作正在 从封闭走向开放 迈出了很大的一步
显示更多
笑死,一个朋友嫌 Seedance 做视频太贵。🤣 转头花了十几万,组装了一台 RTX 5090 机器,专门在本地跑 MiniMax H3 视频工作流。 今天跟我聊天,我给他看了下 Veo + Fable 5.2 做出来的 AI 视频。 他看完沉默了半天: “怎么感觉比我本地跑的效果还好……” 本来想省点 AI 视频生成费,结果直接把未来几年的会员费一次性交了。🤡
显示更多
0
45
81
6
转发到社区
MiniMax 开源的生图模型 MiniMax-H3-Image,原版 68.65GB。几天里社区做出 27 个量化版,Q4 是 10.78G,最小的 Q2_K 压到 6.4G,原生就能出 4096×4096 的图。 你的显卡几个 G,够跑哪一版?
显示更多
昨天最炸裂的消息就是,前OpenAI研究员 @willdepue 发现DeepSeek、Kimi、MiniMax、蚂蚁最新的 6 个开源模型,训练用的全是英伟达 Blackwell。 这款卡是美国明确禁售给中国的,但现在中国的模型训练却用了禁售的卡,卡脖子不完全起作用。 报告估算 5 次训练合计约 12,000 张,按 7 万美元一张,光卡就是 8.4 亿美元。 他怎么看出来的: 1、模型开训前,要先往几十亿个格子里填随机数,填的顺序跟显卡有多少个计算核心绑死。 2、训练完,总有一些格子没被动过,还留着当初的随机数。 3、拿这些残留倒推,H100 那一代读出来是 132,Blackwell 读出来是 148 或 152。 同一批公司更早的模型,读出来还是 H100 那一代。换卡集中在 2026 年 4 月到 10 月,这恰恰也是中国开源模型开始逐渐追上美国顶尖大模型的时间。 根据报告预估训练用的英伟达 Blackwell 训练 - Kimi K3 约 4,500 张 - DeepSeek-V4-Pro 约 3,000 张 - 另外 3 个模型各约 1,500 张 这个数字只是估算不代表真实数据,也不能说明训练用的都是英伟达的卡,但对大模型公司来说时间就是金钱,训练时间短一半,卡数就翻一倍。 kimi K3能在时间和参数量上领先其他国模可能也是因为卡更多的原因 最后需要说明的是,指纹只证明开训用的是什么卡,看不出卡在哪、归谁,海外租的也有可能,之前就有报道说过kimi用海外的算力来训练这在当时并不违规,美国没有明文禁止中国公司在海外租用算力训练模型 报告自己也说不能据此认定走私。DeepSeek 最新的 V4.1-Flash 没找到指纹,Qwen系列也没有被点名,这是真实力。 禁令管住了卖卡,没管住租卡,美国可能在未来补上这个口子。
显示更多
目前本地部署的 AI 服务栈基本凑齐了,链路如下: 🎙️ ASR:QwenASR + Whisper 🗣️ TTS:IndexTTS + BreezyTTS 🎵 音乐:Yue2 + MiniMax Music 🖼️ 图像:QwenImage 2.1 🧠 LLM:Qwen3.8 27B 🎬 视频:MiniMax H3 + LTX 2.5 从语音识别、语音合成、音乐、图像、大模型到视频生成,本地自建服务一条龙
显示更多
0
9
275
30
转发到社区
68.65 GB 硬生生砍到 6.4 GB,开源社区又把大厂生图门槛给干碎了! MiniMax 刚开源的 33.1B 生图核弹 MiniMax-H3-Image,原版直接劝退 99% 的家用电脑,结果几天内被社区榨出 27 个量化版,狂下超 11 万次! 实测爽点极其凶残: · 显存门槛大跳水:原版 68G 连 4090 都喘,现在 Q4 仅 10.78G,极限 Q2_K 直接压到 6.4G!连轻薄 Mac 都能强行跑,Unsloth FP8(18.8G)直接拉满画质 · 告别放大器,原生直出 4K:以前跑 4K 图显存动不动当场爆掉,它原生直接硬吞 4096×4096(1600万像素),甚至连 21:9 的电影宽屏大图都不用二道放大 · 6 步极速秒出:官方蒸馏到 6 步采样直接成图,不需要开 CFG,告别跑一张图等半天的便秘感 · 生图+修图一体通吃:支持喂进 16 张参考图定控角色,ComfyUI GGUF、Mac MLX 现成权重全部就绪,挑好显存直接下 手里只有 8G、12G 显卡,又馋大厂 4K 生图画质的兄弟,这波现成作业必须抄。 量化合集入口: 你平时在本地跑生图,最怕显存直接爆红还是出图慢得像便秘? #AI# #开源# #ComfyUI#
显示更多
0
23
636
66
转发到社区
卧槽,minimax H3生图模型屌爆了! 直接生成千万像素图片 完爆QWEN生图。 而且有些模型无审查,尺度大。 测试了下居然能出黄果图。 这下黄果视频彻底没门槛了,很刑。
显示更多
MiniMax H3-Image is the #1# open model in Multi-Image Edit Arena, scoring 1365 (#11# overall) and tied with Nano Banana 2 and Nano Banana Pro. It is the only open model in the top 20. It also lands: - #2# open model in Single-Image Edit (1359, tied with Qwen-Image-2.1) - #3# open in Text-to-Image (1197, tied with Ideogram 4.0) The lab reports this as the fastest 2K image editing model: 3x faster than Nano Banana 2.1 at 10-15s of generation time for 2K editing. Open weights are starting to close the gap towards the frontier. Congrats to the @minimax_AI team!
显示更多
0
19
87
2
转发到社区
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
显示更多
0
1.7K
38.5K
5.9K
转发到社区
分享 几段不错的下腰 h3提示词写法。 | 💛 荧|性感舞蹈 · 柔韧下腰 旅行者荧的舞蹈时间!✨ 随着音乐节奏舒展身体,一个流畅的下腰动作,将柔韧性与舞蹈张力展现得淋漓尽致。优雅中带着一点小小的性感,每个动作都充满镜头表现力。💃 #荧# #Lumine# #原神# #GenshinImpact# #蛍# #原神舞蹈# #性感舞蹈# #下腰# #柔韧舞蹈# #舞蹈展示# #角色舞蹈# #Dance# #DanceVideo# #GenshinDance# #原神コスプレ# #踊ってみた# #루미네# #원신# #H3# #MiniMaxH3#
显示更多
0
10
166
13
转发到社区