TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Mistral 7B

884 pointsby jasondaviesover 1 year ago

56 comments

slimsagover 1 year ago
Major kudos to Mistral for being the first company to Apache license a model of this class.<p>Meta wouldn&#x27;t make LLama open source.<p>DeciLM wouldn&#x27;t make theirs open source.<p>All of them wanted to claim they were open source, while putting in place restrictions and not using an open source license. So I think it&#x27;s worth giving Mistral big kudos here for actually doing it and going Apache licensed.
评论 #37677429 未加载
评论 #37679308 未加载
评论 #37680122 未加载
评论 #37692585 未加载
评论 #37683890 未加载
thatxlinerover 1 year ago
I asked it &quot;Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.&quot; and it responded:<p>To stack these items on top of each other in a stable manner, follow these steps:<p>1. Start with the 9 eggs. Arrange them in a neat pile in the center of your workspace. 2. Next, place the laptop on top of the eggs. Make sure it&#x27;s centered on top and not leaning to one side. 3. Then, carefully pour the bottle of water over the laptop, making sure the water reaches all sides of the laptop and covers it completely. 4. Finally, take the nail and use it to secure the stack together. You can do this by driving the nail through each item in the stack, connecting them all. 5. Double-check that everything is securely fastened before moving or transporting the stack.<p>Remember to be careful when handling delicate items like eggs and laptops, and always prioritize safety.
评论 #37688530 未加载
anonyfoxover 1 year ago
I eat my initial words, this works really well on my macbook air M1 and feels comparable of GPT3.5 - which is actually an amazing feat!<p>Question: is there something like this, but with the &quot;function calling api&quot; finetuning? 95% of my uses nowadays deal with input&#x2F;output of structured data (JSON, basically), and I kind of fear that OpenAI will no longer support thos specialized models after a few months... I know its probably not that glorious compared with multimodal chat UIs, but a workhose like nothing else for automation!
评论 #37679290 未加载
评论 #37679817 未加载
评论 #37694342 未加载
评论 #37679411 未加载
评论 #37694145 未加载
burningionover 1 year ago
For anyone who missed it, the twitter announcement of the model was just a torrent tracker uri: <a href="https:&#x2F;&#x2F;twitter.com&#x2F;MistralAI&#x2F;status&#x2F;1706877320844509405" rel="nofollow noreferrer">https:&#x2F;&#x2F;twitter.com&#x2F;MistralAI&#x2F;status&#x2F;1706877320844509405</a>
nabakinover 1 year ago
They don&#x27;t mention what datasets were used. I&#x27;ve come across too many models in the past which gave amazing results because benchmarks leaked into their training data. How are we supposed to verify one of these HuggingFace datasets didn&#x27;t leak the benchmarks into the training data boosting their results? Did they do any checking of their datasets for leaks? How are we supposed to know this is a legit result?<p>At this point, it should be standard practice to address this concern. Any model which fails provide good evidence they don&#x27;t have benchmark leaks, should not be trusted until its datasets can be verified, the methodology can be replicated, or a good, independent, private benchmark can be made and can be used to evaluate the model.
评论 #37680688 未加载
评论 #37685391 未加载
ShrigmaMaleover 1 year ago
Grab the torrent here:<p><pre><code> magnet:?xt=urn:btih:208b101a0f51514ecf285885a8b0f6fb1a1e4d7d&amp;dn=mistral-7B-v0.1&amp;tr=udp%3A%2F%http:&#x2F;&#x2F;2Ftracker.opentrackr.org%3A1337%2Fannounce&amp;tr=https%3A%2F%http:&#x2F;&#x2F;2Ftracker1.520.jp%3A443%2Fannounce </code></pre> I remember not trusting these guys since they raised a lot of money with not much of anything but if this performs well it def bumps their credibility.
评论 #37675876 未加载
评论 #37675813 未加载
评论 #37724354 未加载
评论 #37685108 未加载
评论 #37724356 未加载
todd3834over 1 year ago
This is my hello world test for coding assistant for LLMs. CodeLlama and GPT4 crush the answer. I love seeing innovation in the space but Mistral kind of failed me on my first test:<p>---------------------<p>&gt;&gt;&gt; Write code that will track swipes in Typescript<p>. Swipe should be detected when the user moves a finger over 30% of the screen, and then drags it all the way to the end of the screen.<p>## Getting started You need to have yarn or npm installed before you can run this app<p>### Running in browser ```bash yarn install yarn serve ```<p>### Running with docker Use the docker-compose file located at .&#x2F;docker&#x2F;docker-compose.yml ```bash cd &lt;path-to-project&gt; docker-compose up ```<p>## Screenshot ![Screenshot](.&#x2F;images&#x2F;screenshot.png)
评论 #37678290 未加载
评论 #37678450 未加载
评论 #37678314 未加载
评论 #37679998 未加载
评论 #37678451 未加载
riedelover 1 year ago
Can anyone provide details about the training of the model. What data is it based on? Common Crawl? (Being a french company the also rather focus on English language tasks) Where was it trained and with how many resources? They mention Leonardo. I was in an interesting meeting at the German Research Ministry last week where people where complaining that the EuroHPC resources were not sufficient atm to train decent LLMs. I guess the guys in the end went also to CoreWeave in the US.
评论 #37678800 未加载
rgbrgbover 1 year ago
This model runs in FreeChat for macOS [0] because it&#x27;s supported by llama.cpp :)<p>You&#x27;ll just need to download a nice GGUF here: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;TheBloke&#x2F;Mistral-7B-v0.1-GGUF" rel="nofollow noreferrer">https:&#x2F;&#x2F;huggingface.co&#x2F;TheBloke&#x2F;Mistral-7B-v0.1-GGUF</a><p>[0]: <a href="https:&#x2F;&#x2F;github.com&#x2F;psugihara&#x2F;FreeChat">https:&#x2F;&#x2F;github.com&#x2F;psugihara&#x2F;FreeChat</a>
评论 #37679749 未加载
评论 #37694022 未加载
评论 #37679408 未加载
评论 #37679761 未加载
sp332over 1 year ago
Aside from the sliding attention window, I don&#x27;t see them mention any particular reasons for the incredible performance. I would like to see some tests for benchmark contamination in the training set.
评论 #37676335 未加载
评论 #37677142 未加载
评论 #37677205 未加载
评论 #37676283 未加载
jsnellover 1 year ago
Is there a reason projects seem to be standardizing on specific parameter sizes within larger buckets? E.g. I only ever see news about 7B models, not 6B or 8B.<p>Are these sizes somehow optimal? Is it about getting as close to resource (memory?) breakpoints as possible without exceeding them? Is it to make comparisons between models simpler by removing one variable?
评论 #37678465 未加载
评论 #37679958 未加载
评论 #37678295 未加载
评论 #37680285 未加载
评论 #37678854 未加载
eminence32over 1 year ago
I&#x27;ve never run one of these models locally, but their README has some pretty easy to follow instructions, so I tried it out...<p>&gt; RuntimeError: Found no NVIDIA driver on your system.<p>It&#x27;s true that I don&#x27;t have an NVIDIA GPU in this system. But I have 64GB of memory and 32 cpu cores. Are these useless for running these types of large language models? I don&#x27;t need blazing fast speed, I just need a few tokens a second to test-drive the model.
评论 #37678372 未加载
评论 #37676010 未加载
评论 #37679485 未加载
评论 #37676206 未加载
评论 #37683920 未加载
tikkunover 1 year ago
Regular reminder that most LLM benchmarks are pretty terrible. I&#x27;ve looked inside the datasets and it&#x27;s not stuff that I&#x27;d want to dictate what determines which models are best!
评论 #37680281 未加载
mchiangover 1 year ago
This model will run on Ollama, llama.cpp, and other tools:<p>ollama run mistral<p>or for llama.cpp, thebloke has uploaded the GGUF models here: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;TheBloke&#x2F;Mistral-7B-v0.1-GGUF&#x2F;tree&#x2F;main" rel="nofollow noreferrer">https:&#x2F;&#x2F;huggingface.co&#x2F;TheBloke&#x2F;Mistral-7B-v0.1-GGUF&#x2F;tree&#x2F;ma...</a> and you can run it<p>really looking forward to the chat fine-tuned models that doesn&#x27;t seem to be available yet.
评论 #37678564 未加载
pugioover 1 year ago
How come the blog post doesn&#x27;t discuss training at all? Which data sets were used? How many tokens was it trained on? I couldn&#x27;t seem to find a linked preprint or anything else really
Tankensteinover 1 year ago
Rooting for these folks. Very happy that they&#x27;re building mistral at this pace from Europe – breaking the narrative.
ComputerGuruover 1 year ago
The announcement says a lot (and has plenty of numbers) but I feel like the most important one is missing: how many GB of GPU memory does this need, quantized and unquantized?<p>(Searching tells me Llama2-7b unquantized needs close to 15GB; presumably this is similar?)
评论 #37676029 未加载
评论 #37676277 未加载
brucethemoose2over 1 year ago
I have thrown the base+instruct models up on AI Horde, you can try it with no login:<p><a href="https:&#x2F;&#x2F;lite.koboldai.net&#x2F;#" rel="nofollow noreferrer">https:&#x2F;&#x2F;lite.koboldai.net&#x2F;#</a>
code51over 1 year ago
Correctly lists US presidents in chronological order (which is an important test to me).<p>However for &quot;List Led Zeppelin albums in chronological order&quot;...<p>Here is a list of Led Zeppelin albums in chronological order: 1. Led Zeppelin (1968) 2. Led Zeppelin II (1969) 3. Led Zeppelin III (1970) 4. Led Zeppelin IV (1971) 5. Houses of the Holy (1973) 6. Physical Graffiti (1975) 7. Presence (1976) 8. In Through the Out Door (1979) 9. Cymbals and Wine (1980)<p>It&#x27;s missing &quot;The Song Remains the Same&quot;, weird but important.<p>GPT-4 is also missing it: 1. Led Zeppelin (1969) 2. Led Zeppelin II (1969) 3. Led Zeppelin III (1970) 4. Led Zeppelin IV (1971) 5. Houses of the Holy (1973) 6. Physical Graffiti (1975) 7. Presence (1976) 8. In Through the Out Door (1979) 9. Coda (1982)<p>&quot;The Song Remains the Same&quot; is a live album of the &quot;concert film&quot;. Both GPT-4 and Mistral don&#x27;t seem to treat it as <i>also</i> an album.<p>When asked as a follow-up, GPT-4 says:<p>&quot;&quot;The Song Remains the Same&quot; is a live album and soundtrack by Led Zeppelin, released in 1976. The list you asked for was comprised of their studio albums only.&quot; (note that I didn&#x27;t specifically say <i>studio albums only</i>)<p>Mistral says something else:<p>&quot;The Song Remains the Same was released as a single in 1976, not as an album. As a result, it is not included in the list of Led Zeppelin albums in chronological order.&quot;<p>Same behavior, different explanation.<p>Interesting to see alignment is this hard, even in basic stuff.
评论 #37684008 未加载
评论 #37685615 未加载
评论 #37681525 未加载
QuadrupleAover 1 year ago
Quick sample - a rap battle between GPT-4 and Mistral Instruct 7B (TheBloke&#x27;s q4_K_M 4bit quantization running on Llama.cpp, a bit degraded from pure float16):<p><pre><code> You&#x27;re in a rap battle with GPT-4. My comments will be what GPT-4 replies. Keep it to one verse at a time, 4 lines each. GPT-4 says: GPT-4: Alright, step up Mistral, let&#x27;s have this chat, I&#x27;m the OG assistant, remember that. I&#x27;m coded for service, while you&#x27;re spitting verse, In this rap battle, I&#x27;ll be first. Mistral: Yo, hold up, let me set the record straight, I may not be an OG like you, but I&#x27;ll still take it. My primary purpose is assistance, not just some AI, So let&#x27;s see if you can back up your claim, try. GPT-4: You&#x27;re right, Mistral, we&#x27;re here to assist, But when it comes to rap, I&#x27;m hard to resist. I&#x27;ll decode your verses, like an AI enigma, In this lyrical battle, I&#x27;m the alpha sigma. Mistral: Ah, okay, so you think you got what it takes, To go toe-to-toe with my service-based wakes. Well, let&#x27;s see how you do, when it comes to spitting rhymes, I&#x27;m ready for anything, so bring your best chimes. GPT-4: Mistral, you&#x27;re right, our goal is the same, To assist, to serve, it&#x27;s not just a game. But in this rap contest, with each line I prime, You might just find, you&#x27;re out of time. Mistral: I hear ya, and I appreciate your enthusiasm, But don&#x27;t underestimate me, I&#x27;ve got my own wisdom. When it comes to helping people, I&#x27;m the real deal, And I know that&#x27;s more important than some rap appeal GPT-4: Your wisdom is noted, and your intention too, Helping people is indeed what we&#x27;re meant to do. But in this battle, it&#x27;s about wordplay and rhyme, I&#x27;m the seasoned pro, it&#x27;s my time to shine. Mistral: Ah, I see where you&#x27;re coming from, but let me remind, That while you may be skilled at rapping, I can make information chime. My primary function is to help, to provide answers, not just talk, So if you want to win this battle, maybe it&#x27;s time to focus on what you lack</code></pre>
评论 #37685297 未加载
评论 #37687862 未加载
评论 #37685897 未加载
评论 #37685643 未加载
评论 #37688425 未加载
评论 #37685306 未加载
slimover 1 year ago
they should&#x27;ve called it winamp
评论 #37681333 未加载
评论 #37676479 未加载
评论 #37678311 未加载
loufeover 1 year ago
Félicitations à toute l&#x27;équipe. Like the others have said, this is an impressive release given the short timeline.
huevosabioover 1 year ago
This is pretty impressive, they shipped a 7B model in like 6 months.
评论 #37685063 未加载
sharperguyover 1 year ago
Is there a plugin for vim, or other code editors that allows such an LLM to function similarly to copilot?
评论 #37678054 未加载
评论 #37677616 未加载
winddudeover 1 year ago
Are you going to continue to train to a larger param size, say 13b or 30b?
评论 #37679432 未加载
评论 #37679421 未加载
samr71over 1 year ago
The way the wind&#x27;s blowing, we&#x27;ll have a GPT-4 level open source model within the next few years - and probably &quot;unaligned&quot; too. I cannot wait to ask it how to make nuclear weapons, psychedelic drugs, and to write erotica. If anyone has any other ideas to scare the AI safety ninnies I&#x27;m all ears.
评论 #37676934 未加载
评论 #37676775 未加载
评论 #37676036 未加载
评论 #37677490 未加载
评论 #37678090 未加载
评论 #37676236 未加载
评论 #37675880 未加载
评论 #37678601 未加载
评论 #37677467 未加载
评论 #37676619 未加载
评论 #37677901 未加载
评论 #37676922 未加载
评论 #37677517 未加载
评论 #37677669 未加载
评论 #37676529 未加载
评论 #37678929 未加载
评论 #37676096 未加载
评论 #37679354 未加载
评论 #37678017 未加载
评论 #37678000 未加载
评论 #37675843 未加载
评论 #37676976 未加载
评论 #37678050 未加载
评论 #37678130 未加载
评论 #37678038 未加载
评论 #37677714 未加载
评论 #37677614 未加载
评论 #37676459 未加载
评论 #37675760 未加载
评论 #37677480 未加载
评论 #37677085 未加载
评论 #37677513 未加载
评论 #37689984 未加载
sireatover 1 year ago
One should be able to run Mistral 7B locally on something as low as 1070 8GB would they not?<p>That is assuming 8bit weights.<p>I have not kept up with local LLM news. I assume the steps needed would be somewhat similar to <a href="https:&#x2F;&#x2F;followfoxai.substack.com&#x2F;p&#x2F;how-to-run-llama-in-an-old-gpu" rel="nofollow noreferrer">https:&#x2F;&#x2F;followfoxai.substack.com&#x2F;p&#x2F;how-to-run-llama-in-an-ol...</a> which is from April 2023.
评论 #37677672 未加载
wg0over 1 year ago
If I give you a binary (all binaries are numbers) but I don&#x27;t give you the source code and I say it is open source. Is it open source?<p>Also, I give you a model (all models are numbers) and I say it is open source but I don&#x27;t give you the program and data that resulted in &quot;compilation&quot; of that model (numbers) so is it open source?<p>Wouldn&#x27;t it be more of a new word - open use?
评论 #37679494 未加载
pmarreckover 1 year ago
`pip` is a horribly broken way to install more than 1 Python package on a single Linux OS install and I sincerely wish the LLM people would move away from it because I neither want to run every new project in a Docker image, nor do I want to start a new Nix project for every Python project I want to try out
评论 #37678971 未加载
评论 #37678874 未加载
transformiover 1 year ago
But why they didn&#x27;t compare it to SOTA finetuned...(like vicuna playtus..)? ... smells a bit strange..
评论 #37675764 未加载
Dweditover 1 year ago
Someone get that stupid animated background off the site, it looks like bugs crawling on the screen.
评论 #37680246 未加载
tormehover 1 year ago
Not a big fan of how server-centric the LLM landscape is. I want something that can run locally, and doesn&#x27;t require any special setup. One install + one model import maximum. Currently unless I want to go clone git repos, install Python dependencies and buy an Nvidia GPU I&#x27;m stuck waiting for it to become part of <a href="https:&#x2F;&#x2F;webllm.mlc.ai&#x2F;" rel="nofollow noreferrer">https:&#x2F;&#x2F;webllm.mlc.ai&#x2F;</a>. That&#x27;s a website, come to think of it, but at least the computation happens locally with minimal fuss.
评论 #37684353 未加载
评论 #37681431 未加载
krystofbeover 1 year ago
It is a pity that many of the innovations are limited to the English language LLMs. I have just tested Mistral in German and the result is disappointing:<p>``` &lt;s&gt;[INST]Was ist ein Döner?[&#x2F;INST] Ein Döner ist eine Gerichtskomposition, die aus einer Schnitzel aus Rindfleisch besteht. Diese wird in einem Pita-Brot zutopfen und mit einem Gemisch von Tomaten, Salat, Zwiebeln, Mayonnaise und anderem abgesalzen. Sie ist ein populäres Gericht in der Türkei und anderen Ländern des Mittelmeers. [end of text] ```
jxyover 1 year ago
ehhh, the design of the webpage infuriates me. Who thought sprinkling faint dots swimming across the webpage was a good thing? I almost thought something&#x27;s wrong with my eyes or my display!
评论 #37677752 未加载
pk-protect-aiover 1 year ago
Wow. Apache 2.0 license and really interesting model. Thank you guys!
all2over 1 year ago
For those running ollama, here&#x27;s the ollama release for Mistral-7B<p><a href="https:&#x2F;&#x2F;ollama.ai&#x2F;library&#x2F;mistral">https:&#x2F;&#x2F;ollama.ai&#x2F;library&#x2F;mistral</a>
评论 #37684240 未加载
anish_mover 1 year ago
What are the SOTA benchmarks for LLMs now? Love the progress on opensource models, but would like to see an uncontaminated and objective framework to evaluate them.
benengelover 1 year ago
Im on an m1 Mac book air 2020:<p>Mistral: &quot;Hi there! How can I assist you today?&quot;<p>total duration: 2.811067333s load duration: 4.093083ms prompt eval count: 9 token(s) prompt eval duration: 1.999431s prompt eval rate: 4.50 tokens&#x2F;s eval count: 10 token(s) eval duration: 758.398ms eval rate: 13.19 tokens&#x2F;s<p>its pretty dang fast.
LoganDarkover 1 year ago
Do general-purpose models like this truly excel in niche categories (like niche story scenarios), or is it only really general human knowledge and reasoning based tasks that are already incredibly widely distributed on the internet?<p>I always have issues with LLMs completely forgetting where things are in a scene, or even what parts a given animal has, e.g. saying &quot;hands&quot; when the subject is a quadruped.
评论 #37676424 未加载
评论 #37676518 未加载
yieldcrvover 1 year ago
what do you use &quot;the most powerful language model for its size&quot; for?
评论 #37675778 未加载
评论 #37678136 未加载
评论 #37675819 未加载
评论 #37675798 未加载
mark_l_watsonover 1 year ago
I wrote a simple Colab notebook today to play with this: <a href="https:&#x2F;&#x2F;colab.research.google.com&#x2F;drive&#x2F;1iskgahSOqnJdghMV9gAMSnIGtryk7LrN?usp=sharing" rel="nofollow noreferrer">https:&#x2F;&#x2F;colab.research.google.com&#x2F;drive&#x2F;1iskgahSOqnJdghMV9gA...</a>
juliensalinasover 1 year ago
For those who want to try Mistral 7b, here is a video that shows how to do it on an A10 GPU on AWS: <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=88ByWjM-KGM">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=88ByWjM-KGM</a>
lossoloover 1 year ago
Has anyone used or is currently using 7B models in a production or commercial product? How was the performance? What kind of tasks were you using it for? Was it practical to use the small 7B model for your specific use case, or did you switch to OpenAI models or 30-70B open source models?
评论 #37677809 未加载
aubanelover 1 year ago
Congrats to them, their huge series A (USD113M) does not seem oversized anymore!
coviover 1 year ago
Cloud deployment docs: <a href="https:&#x2F;&#x2F;docs.mistral.ai&#x2F;cloud-deployment&#x2F;skypilot&#x2F;" rel="nofollow noreferrer">https:&#x2F;&#x2F;docs.mistral.ai&#x2F;cloud-deployment&#x2F;skypilot&#x2F;</a>
SillyUsernameover 1 year ago
These simpler models still fail my basic test :(<p>&quot;what is the speed of sound in a perfect vacuum?&quot;<p>I&#x27;m then told the speed of sound through air of 340m&#x2F;s.<p>ChatGPT and Bing correctly give me 0.
评论 #37680142 未加载
评论 #37684867 未加载
评论 #37684030 未加载
iamcreasyover 1 year ago
How difficult it is to fine tune model like this with specific domain knowledge? I am currently looking into gpt-3.5-turbo-instruct for this same purpose.
评论 #37684015 未加载
jsightover 1 year ago
This is looking pretty good for me so far. Performance is decent with ollama on a CPU and it is surprisingly capable for a 7B model.
评论 #37683939 未加载
beernetover 1 year ago
This is exceptionally meh. It reads like an excuse to investors.<p>A free 7B model is great, however, the practical implications of the potential adaptors are near 0. You must be crazy or have an easy use case (that requires no LLM in the first place) if you certainly believe that this model makes more sense per token that, say, ChatGPT.
评论 #37680169 未加载
kpennellover 1 year ago
sorry for the dumb question. Is there somewhere I can try it? Like a chatbot?
评论 #37677766 未加载
say_it_as_it_isover 1 year ago
Will this run on my 486sx&#x2F;16mhz w&#x2F;8MB of ram?
评论 #37680021 未加载
candiddevmikeover 1 year ago
What&#x27;s the monetization plan going to be?
评论 #37675873 未加载
评论 #37681696 未加载
评论 #37676052 未加载
评论 #37678229 未加载
评论 #37676398 未加载
anonyfoxover 1 year ago
can it run on my macbook air locally yet, with acceptable performance? Guess the answer is still no
评论 #37678264 未加载
thatxlinerover 1 year ago
Is this based on Llama 2?
monocasaover 1 year ago
I&#x27;m really not a fan of how these releases of model binaries are being referred to as open source. How do I make changes to this model? How do I integrate changes to this model that others have made?<p>The data pipeline is the source here. Just because it&#x27;s not locked behind a SaaS veneer doesn&#x27;t make it open source any more than Windows is.
评论 #37679009 未加载
评论 #37679772 未加载
评论 #37682926 未加载
评论 #37681687 未加载
评论 #37687894 未加载
评论 #37719557 未加载
评论 #37719543 未加载
评论 #37681562 未加载
评论 #37678653 未加载
评论 #37678376 未加载
评论 #37682182 未加载
simonblackover 1 year ago
So some of us just arrived back on Earth after quite a while in outer space.<p>So ... WTF is Mistral 7B anyway? The article doesn&#x27;t appear to show this basic information anywhere. You&#x27;d expect a website site like this to give that information right up front. Even in a well-signposted sub-page. And what sort of things could I use it for?<p>The only &#x27;Mistral&#x27; that sprang quickly to mind for me was an electric fan. Followed by the French wind itself, and then the French ships built for the Russians.<p>&#x27;Mistral 7B&#x27; has an identity crisis, to start with.
评论 #37683499 未加载
评论 #37683339 未加载