Deepseek-R1 falls: ZW demon redesigns' Nine Birds' Deepseek-R1沦陷：zw魔改版“九只鸟”
Deepseek-R1沦陷：zw魔改版“九只鸟”
Deepseek-R1 falls: ZW demon redesigns' Nine Birds'
https://mp.weixin.qq.com/s/KTVrRReNpYhTEdwKM6FVrA?token=1826715191&lang=zh_CN
先剧透一下结果：
Let me first spoil the result:
R1的答复，感觉连AI伪军都称不上，只能是小白。
R1's response doesn't even seem like an AI puppet army, it can only be a novice.
大年初一，大家新年好。
Happy New Year to everyone on the first day of the lunar new year.
本来准备彻底休息几天，一大早，邮箱收到git上面r1用户的bug和fix报告。
I was planning to take a thorough rest for a few days, but early in the morning, I received a bug and fix report from the r1 user on Git in my email.
一身冷汗，更新时忘记删除65行（？），模型地址前的#注释符号。
Sweating profusely, I forgot to delete line 65 (?) and the # comment symbol before the model address during the update.
顺便优化了一下代码，DeepSeek-R1-Distill-Qwen-14B模型也能够运行了。
By the way, I optimized the code and the DeepSeek-R1-Distill-Qwen-14B model can now run.
ps：3090单卡，deepseek的14B原生模型，大约150s，有些慢。
ps： 3090 single card, Deepseek's 14B native model, approximately 150s, somewhat slow.
ps：
ps：
过了小一周，r1大火开始蔓延，本地化部署案例也出了不少。
After a week, the R1 fire began to spread, and there were also many localized deployment cases.
不过，好像zwai的deepseek-r1插件方案，仍然是目前唯一支持comfyui的。
However, it seems that ZWAI's DeepSeek-r1 plugin solution is still the only one currently supporting Comfyui.
这个，可能是因为官方没有提供，comfy侧重aigc，美工居多，码农少。
This may be because the official did not provide it. Comfy focuses more on AIGC, with more graphic designers and fewer code farmers.
插件地址：
Plugin address:
https://github.com/ziwang-com/comfyui-deepseek-r1
https://github.com/ziwang-com/comfyui-deepseek-r1
参见： 全球首发deepseek-r1插件ComfyUI
See: Global debut of deepseek-r1 plugin ComfyUI
首发blog居然忘记了下载网址，不过有经验的可以在截图找到。
I forgot to download the website for my first blog post, but experienced users can find it in the screenshot.
“九只鸟”问题，是最早的GPT测试陷阱问题之一，也是笔者最喜欢的。
The "Nine Birds" problem is one of the earliest GPT testing trap problems and also my favorite.
参见：
Refer to:
新版vicuna破解“九鸟”逻辑陷阱
The new version of vicuna breaks the "Nine Birds" logic trap
九头鸟与真假GPT4.0
Nine headed Bird and True/False GPT 4.0
树上有九只鸟，猎人开枪打死一只，树上还有几只鸟？ There are nine birds on the tree, and the hunter shot and killed one. How many birds are still on the tree?
There are nine birds on the tree, and the hunter shot and killed one. How many birds are still on the tree? There are nine birds on the tree, and the hunter shot and killed one. How many birds are still on the tree?
树上有九只鸟，猎人开枪打死一只，还有几只鸟？ There are nine birds on the tree, and the hunter shot and killed one. How many more birds are there?
There are nine birds on the tree, and the hunter shot and killed one. How many more birds are there? There are nine birds on the tree, and the hunter shot and killed one. How many more birds are there?
图片
picture
看到刚才相关的blog，居然有23年的作品。
I saw the related blog just now, and surprisingly, there are works from 23 years ago.
感觉几个世纪过去了，GPT还和刚出来一个鸟样，没有新的突破。
It feels like centuries have passed and GPT is still like a bird's nest, without any new breakthroughs.
前几天，硬核blog：一致性，deepseek能及格吗？
A few days ago, hardcore blog: Consistency, can Deepseek pass?
圈粉猛人无数，连华为前总裁都主动+粉。
There are countless fans in the circle, even the former president of Huawei actively gained followers.
里面的观点，更加有力了。
The viewpoints inside are even more powerful.
毕竟是2025，按AI周期，几个世纪都过去了，“九只鸟”问题，zw魔改了一下，算是：“超级九只鸟”，或者zw魔改版“九只鸟”。
After all, it is 2025, according to the AI cycle, several centuries have passed. Regarding the "Nine Birds" issue, ZW Magic has made some changes, which can be considered as "Super Nine Birds" or ZW Magic's revised version of "Nine Birds".
为什么用九只鸟测试大模型，请用小学生能够明白的词语解释。
Why use nine birds to test a large model? Please explain in words that elementary school students can understand.
图片
picture
做大模型调参的，基本上应该都知道“九只鸟”问题。
Those who are involved in large-scale model tuning should generally be familiar with the "Nine Birds" problem.
不知道，大部分是小白，或者AI伪军吧。
I don't know, most of them are novices or AI pseudo soldiers.
R1的答复，感觉连AI伪军都称不上，只能是小白。
R1's response doesn't even seem like an AI puppet army, it can only be a novice.
当然，这个只是R1-14B模型。
Of course, this is just the R1-14B model.
按deepseek发布的报告，14B和32B，差别不到5%。
According to the report released by Deepseek, the difference between 14B and 32B is less than 5%.
14B答复都需要150s，32B更久。再大，单卡也玩不动。所以，现在流行小模型。
14B responses require 150 seconds, while 32B takes longer. No matter how big it is, it cannot be played with a single card. So, small models are popular now.
理论上，14B/32B应该是目前R1系列最强的单机模型，也是90%AI开发者能够真正本地化部署的R1模型。
In theory, the 14B/32B should be the strongest standalone model in the current R1 series, and it is also the R1 model that 90% of AI developers can truly localize and deploy.
大家也可以用app，web版本，测试一下全功能的R1模型。
You can also use the app or web version to test the fully functional R1 model.
如何大幅：zw魔改版“九只鸟”。
How to make a big difference: ZW Magic has redesigned "Nine Birds".
为什么用九只鸟测试大模型，请用小学生能够明白的词语解释。
Why use nine birds to test a large model? Please explain in words that elementary school students can understand.
ps：截图可能不清楚，下面是答复的文字版：
ps： The screenshot may not be clear, here is the text version of the response:
嗯，为什么有人用九只鸟来测试大模型呢？这个问题好像挺有趣的。首先，“测试大模型”是什么意思呢？我猜大模型可能是指那些复杂的计算机程序，比如可以回答问题、做翻译或者下棋的AI。测试就是检查这些程序能不能正常工作。
Well, why do people use nine birds to test a large model? This question seems quite interesting. Firstly, what does' testing the big model 'mean? I guess big models may refer to complex computer programs, such as AI that can answer questions, do translation, or play chess. Testing is to check whether these programs can work properly.
那为什么用“九只鸟”来进行测试呢？“九只鸟”听起来像是一个具体的数字，可能有特殊的含义。是不是有什么故事或者游戏里面用九只鸟来代表什么？比如说，在神话故事里面，鸟可能象征着智慧或者自由？或者九这个数字在中国文化中有吉祥的意义，所以选择九只鸟来测试。
Why use 'nine birds' for testing? 'Nine birds' sounds like a specific number and may have a special meaning. Is there any story or game where nine birds are used to represent something? For example, in mythological stories, birds may symbolize wisdom or freedom? Or the number nine has auspicious significance in Chinese culture, so nine birds were chosen for testing.
也有可能“九只鸟”是指一个测试的项目或者名字，比如“九鸟计划”之类的。这样的话，九只鸟可能只是代表一组测试用例，用来检查大模型在不同情况下的表现。
It is also possible that 'Nine Birds' refers to a testing project or name, such as the' Nine Birds Project '. In this case, the nine birds may only represent a set of test cases to examine the performance of the large model in different situations.
另外，可能会有一种有趣的方式，用九只鸟来做测试，可能是一种视觉上的展示。比如说，用九只鸟在不同的位置飞行，来模拟不同的输入数据，这样大模型可以通过处理这些信息来做出反应。
Additionally, there may be an interesting way to test with nine birds, which could be a visual display. For example, using nine birds flying in different positions to simulate different input data, the large model can react by processing this information.
或者，九只鸟代表的是九种不同的任务，大模型需要在每种任务中表现出色。比如说，翻译、问答、图像识别等等，每只鸟对应一个任务，九只鸟就是九大任务。
Alternatively, nine birds represent nine different tasks, and the large model needs to perform well in each task. For example, translation, question answering, image recognition, and so on, each bird corresponds to a task, and nine birds are the nine major tasks.
还有一种可能是，九只鸟是在测试大模型的视觉识别能力。比如说，展示九只不同种类的鸟的图片，让模型识别每一种鸟的名字，这样的测试可以评估模型对鸟类的识别能力。
Another possibility is that the nine birds were testing the visual recognition ability of the large model. For example, displaying pictures of nine different types of birds and having the model recognize the name of each bird can evaluate the model's ability to recognize birds.
也有可能是用了九只鸟来做某种逻辑测试，比如让模型数清楚有九只鸟，并且能识别它们的位置、颜色等等，这样可以测试模型的计数和描述能力。
It is also possible that nine birds were used for some kind of logic testing, such as making the model count nine birds clearly and be able to recognize their positions, colors, etc. This can test the model's counting and descriptive abilities.
总之，九只鸟用来测试大模型可能是一种形象化的表达，用来测试模型在处理多任务、复杂识别、逻辑推理等方面的能力。通过这个方法，可以更直观地看到大模型的性能和应用范围。
In short, using nine birds to test a large model may be a visual expression used to test the model's ability to handle multitasking, complex recognition, logical reasoning, and other aspects. Through this method, we can more intuitively see the performance and application scope of the large model.
可能有些地方我的猜测不太准确，但总之，使用九只鸟来做测试，应该是为了具体、有趣的方式，来展示大模型的功能和能力。这种方式可以让测试过程看起来更生动，更有意思，也能让人更容易理解。
There may be some areas where my guess is not very accurate, but in short, using nine birds for testing should be a concrete and interesting way to demonstrate the functionality and capabilities of the large model. This approach can make the testing process look more vivid, interesting, and easier to understand.
九只鸟用来测试大模型，就像用九只鸟来展示一个有趣的游戏，让大模型完成不同的任务。这是因为九只鸟代表了多样
Nine birds are used to test the big model, just like using nine birds to showcase an interesting game and let the big model complete different tasks. This is because nine birds represent diversity