TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Llama-3.3-70B-Instruct

425 点作者 pr337h4m5 个月前

26 条评论

paxys5 个月前
Benchmarks - <a href="https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;LocalLLaMA&#x2F;comments&#x2F;1h85ld5&#x2F;comment&#x2F;m0qauyg&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;LocalLLaMA&#x2F;comments&#x2F;1h85ld5&#x2F;comment...</a><p>Seems to perform on par with or slightly better than Llama 3.2 405B, which is crazy impressive.<p>Edit: According to Zuck (<a href="https:&#x2F;&#x2F;www.instagram.com&#x2F;p&#x2F;DDPm9gqv2cW&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.instagram.com&#x2F;p&#x2F;DDPm9gqv2cW&#x2F;</a>) this is the last release in the Llama 3 series, and we&#x27;ll see Llama 4 in 2025. Hype!!
评论 #42343012 未加载
评论 #42341998 未加载
评论 #42342566 未加载
ben305 个月前
This reminds me of Steve Jobs&#x27;s famous comment to Dropbox about storage being &#x27;a feature, not a product.&#x27; Zuckerberg - by open-sourcing these powerful models, he&#x27;s effectively commoditising AI while Meta&#x27;s real business model remains centred around their social platforms. They can leverage these models to enhance Facebook and Instagram&#x27;s services while simultaneously benefiting from the community improvements and attention. It&#x27;s not about selling AI; it&#x27;s about using AI to strengthen their core business. By making it open, they get the benefits of widespread adoption and development without needing to monetise the models directly.
评论 #42343031 未加载
评论 #42343254 未加载
评论 #42344139 未加载
评论 #42343397 未加载
评论 #42346371 未加载
评论 #42343585 未加载
评论 #42343303 未加载
评论 #42344013 未加载
LorenDB5 个月前
Seems to be more or less on par with GPT-4o across many benchmarks: <a href="https:&#x2F;&#x2F;x.com&#x2F;Ahmad_Al_Dahle&#x2F;status&#x2F;1865071436630778109" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;Ahmad_Al_Dahle&#x2F;status&#x2F;1865071436630778109</a>
评论 #42342258 未加载
评论 #42344363 未加载
评论 #42342356 未加载
评论 #42342588 未加载
freediver5 个月前
Does unexpectedly well on our benchmark:<p><a href="https:&#x2F;&#x2F;help.kagi.com&#x2F;kagi&#x2F;ai&#x2F;llm-benchmark.html" rel="nofollow">https:&#x2F;&#x2F;help.kagi.com&#x2F;kagi&#x2F;ai&#x2F;llm-benchmark.html</a><p>Will dive into it more, but this is impressive.
评论 #42346793 未加载
profsummergig5 个月前
Please help me understand something.<p>I&#x27;ve been out of the loop with HuggingFace models.<p>What can you do with these models?<p>1. Can you download them and run them on your Laptop via JupyterLab?<p>2. What benefits does that get you?<p>3. Can you update them regularly (with new data on the internet, e.g.)?<p>4. Can you finetune them for a specific use case (e.g. GeoSpatial data)?<p>5. How difficult and time-consuming (person-hours) is it to finetune a model?<p>(If HuggingFace has answers to these questions, please point me to the URL. HuggingFace, to me, seems like the early days of GitHub. A small number were heavy users, but the rest were left scratching their heads and wondering how to use it.)<p>Granted it&#x27;s a newbie question, but answers will be beneficial to a lot of us out there.
评论 #42342569 未加载
评论 #42342540 未加载
评论 #42343625 未加载
theanonymousone5 个月前
I&#x27;m &quot;tracking&quot; the price of if 1M tokens in OpenRouter and it is decreasing every few refreshes. It&#x27;s funny: <a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;meta-llama&#x2F;llama-3.3-70b-instruct" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;meta-llama&#x2F;llama-3.3-70b-instruct</a>
danielhanchen5 个月前
I uploaded 4bit bitsandbytes, GGUFs and original 16bit weights to <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth</a> for those interested! You can also finetune Llama 3.3 70B in under 48GB of VRAM and 2x faster and use 70% less memory with Unsloth!
bnchrch5 个月前
Open Sourcing Llama is one of the best example and roll out of &quot;Commoditize Your Complement&quot; in memory.<p>Link to Gwern&#x27;s &quot;Laws of Tech: Commoditize Your Complement&quot; for those who havent heard of this strategy before<p><a href="https:&#x2F;&#x2F;gwern.net&#x2F;complement" rel="nofollow">https:&#x2F;&#x2F;gwern.net&#x2F;complement</a>
评论 #42345482 未加载
hubraumhugo5 个月前
Meta continues to overdeliver. Their goal from the start was to target and disrupt OpenAI&#x2F;Anthropic with a scorched earth approach by releasing powerful open models.<p>The big winners: we developers.
philipkiely5 个月前
Just spent a few minutes this morning spinning up a H100 model server and trying an FP8 quantized version (including kv cache quantization) to fit it on 2 H100s -- speed and quality looking promising.<p>I&#x27;m excited to see if the better instruction following benchmarks improves function calling &#x2F; agentic capabilities.
kstrauser5 个月前
I know this has been discussed before but it changes frequently: what’s the good “generic” Mac desktop client these days? I’d like to use Ollama and&#x2F;or ChatGPT. Maybe Claude. Perhaps Perplexity, too. I primarily want to use AI chats in various apps, like typing “write a function to…” into whatever random editor I’m using at the moment. It doesn’t have to be a desktop app, either. If there’s a great PopClip plugin or Keyboard Maestro macro, or even something that works as a system service, that’s perfectly fine by me.<p>MacMind is nifty, but that feels like a lot of money for something that’s a front end to someone else’s API. “Stop being a cheapskate” is a legitimate answer.
评论 #42347082 未加载
评论 #42346030 未加载
评论 #42348397 未加载
hrpnk5 个月前
Seems that a bunch of quantized models are already uploaded to ollama: <a href="https:&#x2F;&#x2F;ollama.com&#x2F;library&#x2F;llama3.3&#x2F;tags">https:&#x2F;&#x2F;ollama.com&#x2F;library&#x2F;llama3.3&#x2F;tags</a>
adt5 个月前
Model card: <a href="https:&#x2F;&#x2F;github.com&#x2F;meta-llama&#x2F;llama-models&#x2F;blob&#x2F;main&#x2F;models&#x2F;llama3_3&#x2F;MODEL_CARD.md">https:&#x2F;&#x2F;github.com&#x2F;meta-llama&#x2F;llama-models&#x2F;blob&#x2F;main&#x2F;models&#x2F;...</a><p>On the Models Table: <a href="https:&#x2F;&#x2F;lifearchitect.ai&#x2F;models-table&#x2F;" rel="nofollow">https:&#x2F;&#x2F;lifearchitect.ai&#x2F;models-table&#x2F;</a>
LorenDB5 个月前
Hopefully this lands on Groq soon!
评论 #42342621 未加载
评论 #42342064 未加载
theanonymousone5 个月前
Given the comments saying it&#x27;s performance seems comparable to 4o&#x2F;4o-mini, is it safe to say that GPT-4 performance can be achieved with less than 100B parameters,in contrary to what previously was thought?
andy_ppp5 个月前
How many tokens per second can I get on an M4 Max with 128gb of RAM?
评论 #42346153 未加载
评论 #42346025 未加载
ndr_5 个月前
It&#x27;s available on IBM WatsonX, but the Prompt Lab may still report &quot;model unavailable&quot;. This is because of overeager guardrails. These can be turned off, but the German translation for this option is broken too: look for &quot;KI-Guardrails auf&quot; in the upper right.
henry20235 个月前
I&#x27;m building a PC just to run inference on this and the QwQ 32B models.<p>Any suggestions on RAM and GPU I should get?
评论 #42344167 未加载
评论 #42343184 未加载
评论 #42342616 未加载
评论 #42352812 未加载
评论 #42342636 未加载
jadbox5 个月前
Would anyone be willing to compress this down to maybe 14b-20b for us on peasant 16gb rigs?
评论 #42342404 未加载
kordlessagain5 个月前
Summary of discussion: <a href="https:&#x2F;&#x2F;claude.site&#x2F;artifacts&#x2F;635d6816-9f60-4545-aeed-54ba180cfd5e" rel="nofollow">https:&#x2F;&#x2F;claude.site&#x2F;artifacts&#x2F;635d6816-9f60-4545-aeed-54ba18...</a>
nxobject5 个月前
I&#x27;m surprised that, out of all of the East Asian languages, they chose Thai to support: do they have a big office there? (I imagine compared to, say, Japanese or (some form of) Mandarin?)
knighthack5 个月前
Given how censored the 3.2 model was, is I&#x27;m looking forward to the abliterated 3.3 version to see if there&#x27;s any significant improvements there that can replace it.
antirez5 个月前
Hot take after trying it a bit. I was not impressed with llama 3.2, but this one, well, it looks like we finally have a very very strong free LLM.
Narciss5 个月前
This is massive, really cool of meta to open source it
评论 #42342354 未加载
ppp9995 个月前
We need more uncensored models
ulam25 个月前
No base model? disappointed.
评论 #42342242 未加载
评论 #42342175 未加载
评论 #42342112 未加载