A-TIER tool
Nemotron 70B — NVIDIA's open-source LLM benchmark leader
100k ft — What is this?
Nemotron 70B — NVIDIA's open-source LLM benchmark leader
50k ft — Why care?
@mattfarmerai flagging NVIDIA's Nemotron 70B as (at the time) the best open-source LLM. One-line pointer, no deeper analysis.
10k ft — Key takeaways
(no takeaways yet)
5k ft — Tools & techniques
Anthropic, Claude, GPT-4, Hugging Face, Llama, Nemotron
(no techniques tagged)
1k ft — Deep-dive / try it
Nemotron 70B is the best open source large language model 🤯 #llm #nvidia #gpt #ai #ainews #gpt4 #anthropic
mariusbatstrand
14w
I got mine currently crudely running on the CMD 🤞😅
Transcript excerpt
Alright, so something crazy happened in AI this week. Nvidia quietly released a new, free, open-source, large-language model that beats GPT-40 and Claude 3.5's Sonnet. Almost everything. It's called Nemotron 70B, and it's a game-changer. Stick around, at the end I'll show you how to use it completely free. The details are actually insane, so they fine-tuned the 70 billion-parameter Llama 3.1 Instruct model, which usually doesn't have a huge impact. But Nvidia's fine-tuned had a significant effect. This model is now number three on the LM Arena leaderboard, beating Claude 3.5's Sonnet and GPT-40. The only models that outperform it are 01 Preview and 01 Mini. Crazy. This is unheard of for a 70B model. Its reasoning abilities are incredible for its size. It had no issues with the how many Rs are in Strawberry question. It also has the best human-like writing I have seen. It writes so much like a human, even AI detectors don't seem to pick up on it. Another feature? Very low censorship. It isn't completely uncensored, but very close. To use it, I recommend using Hugging Face Chat. It's like a free version of ChatGBT, with no usage limits, tons of additional tools, just select Nemotron 70B as the model. Bam. You can even give Nemotron access to the web on Hugging Chat. This is crazy. I can follow for more AI.