• zcrdwx

    preview版本的config```
    hidden_size = 7168
    n_routed_experts = 384
    num_hidden_layers = 61
    num_attention_heads = 128
    ``但是正式版config,疑似是把flash的配置拷贝过来了``
    hidden_size = 4096
    n_routed_experts = 256
    num_hidden_layers = 43
    num_attention_heads = 64

    
    [1] DeepSeek-AI. *DeepSeek-V4-Pro Preview: config.json*. Hugging Face.
    
    huggingface.co
    
    ![图片](https://cdn3.ldstatic.com/optimized/4X/0/4/d/04de1e0b8c52e3bac013042e6853ab5f237bd0c5_2_690x372.jpeg)
    
    config.json · deepseek-ai/DeepSeek-V4-Pro at main
    
    We’re on a journey to advance and democratize artificial intelligence through open source and open science.
    
    [2] DeepSeek-AI. *DeepSeek-V4-Pro-0813: config.json*. Hugging Face.
    
    huggingface.co
    
    ![图片](https://cdn3.ldstatic.com/optimized/4X/0/e/3/0e3bb80b7d955f20aeaf3b1500706f1bd299599b_2_690x372.jpeg)
    
    config.json · deepseek-ai/DeepSeek-V4-Pro-0813 at main
    
    We’re on a journey to advance and democratize artificial intelligence through open source and open science.
    
    [3] DeepSeek-AI. *DeepSeek-V4-Flash-0731: config.json*. Hugging Face.
    
    huggingface.co
    
    ![图片](https://cdn3.ldstatic.com/optimized/4X/7/8/f/78fa7e9f152d6c37a889960214b25ce24d87b284_2_690x372.jpeg)
    
    config.json · deepseek-ai/DeepSeek-V4-Flash-0731 at main
    
    We’re on a journey to advance and democratize artificial intelligence through open source and open science.
    
    [4] DeepSeek-AI. *DeepSeek-V4-Pro-DSpark: config.json*. Hugging Face.
    
    huggingface.co
    
    ![图片](https://cdn3.ldstatic.com/optimized/4X/9/2/b/92ba9d5e7e01eba51c819923506289b89d85871d_2_690x372.jpeg)
    
    config.json · deepseek-ai/DeepSeek-V4-Pro-DSpark at main
    
    We’re on a journey to advance and democratize artificial intelligence through open source and open science.
    主楼
  • cqting

    牢梁被冲晕了?这两天隔三岔五的出事故 给我整好了啊 亿万鲸子没脸说话了要

    #1
  • aeroides

    疑似是把flash的配置拷贝过来了

    说不定真是 flash ,没考错 ^-^

    #2
  • WWT

    离谱 这波草台班子无敌了

    这下Pro真的是flash了

    图片

    #3
  • LLMeme

    通过开源邀请全世界开发者debug是吧

    #4
  • V_Arrow

    ^-^所以用错config居然不会报错?

    #5
  • 地始F兔

    竟然真的上错模型了?传言成真了?

    #6
  • looklook

    难道说deepseek是一个草台班子?? ^-^

    #7
  • 青墨

    所以说?pro还有救吗?这个配置是flash的?

    #8
  • yukaii

    神tm鲸子,小难梁和他的亿万鲸子

    #9
  • 超大的行李箱

    最近招的人多了,肯定实习生干的^-^

    #10
  • BaoPaper

    越来越搞不懂了,所以真的是 DS 官方把模型搞错了?

    #11
  • LLMeme

    吓得官方撤回了

    [image]
    DeepSeek-V4-Pro-0813权重下架

    #12
  • xiaofei_xiao

    (帖子已被作者删除)

    #13
  • role_c

    ds好像没有股价 吧, 怎么跌。

    #14
  • wanmu

    都没上市哪来的股价,佬友喷也得喷好的呀 ^-^

    #15

发表回复

登录后回复