🚨 Cerebras Inference is now 3x faster:
Llama3.1-70B just broke 2,100 tokens/s
- 16x faster than the fastest GPU solution
- 8x faster than GPUs running Llama *3B*
- It's like the perf of a new hardware generation in a single software release
Available now at inference.cerebras.ai We broke all records when we launched Cerebras Inference in August. Today we are tripling our performance from 650 t/s to 2100 t/s.
Cerebras Inference speed is in a league of its own – 16x faster than the fastest GPU solution, 68x faster than hyperscale clouds, and 4-8x faster than other AI accelerators.

Introducing Cerebras Inference
‣ Llama3.1-70B at 450 tokens/s – 20x faster than GPUs
‣ 60c per M tokens – a fifth the price of hyperscalers
‣ Full 16-bit precision for full model accuracy
‣ Generous rate limits for devs
Try now: inference.cerebras.ai Cerebras Inference is the fastest Llama3.1 inference API by far: 1,800 tokens/s for 8B and 450tokens/s for 70B. We are ~20x faster than NVIDA GPUs and ~2x faster than Groq.

Introducing BTLM-3B-8K: an open, state-of-the art 3B parameter model with 7B level performance. When quantized, it fits in as little as 3GB of memory 🤯. It runs on iPhone, Google Pixel, even Raspberry Pi. BTLM goes live on Bittensor later this week! 🧵👇
https://t.co/7aKLkeUeUIbuff.ly/3Q5dtY5
Today's popular models can run on a powerful PC but don't fit in popular mobile devices. In May @Opentensor challenged us to build a SoTA model that runs on any device and supports long context. Thus was born BTLM - a 3B model with 7B performance and 8K context length!

📣 Today we are announcing Condor Galaxy-1: a 4 exaflop AI supercomputer built in partnership with @G42ai. Powered by 64 Cerebras CS-2 systems, 54M cores, and 82TB of memory – it's the largest AI supercomputer we've ever built. But that's not all: CG-1 is just the start.. G42 Cloud is the largest public cloud provider of the UAE. To expand its AI offering, we are planning not one but *nine* AI supercomputers. When complete in 2024, the full Condor Galaxy system will have 9 instances, 576 CS-2s, for a total of 36 exaFLOPs of AI compute. 🤯🤯

📣 New dataset drop!
Introducing SlimPajama-627B: the largest extensively deduplicated, multi-corpora, open-source dataset for training large language models. 🧵cerebras.net/blog/slimpajam… RedPajama-1T is the largest open dataset today but contains a large percentage of duplicates, making a full training run costly and inefficient. Like the Falcon team, we found data quality is just as important as quantity – which led to SlimPajama. huggingface.co/datasets/cereb…

🎉 Exciting news! Today we are releasing Cerebras-GPT, a family of 7 GPT models from 111M to 13B parameters trained using the Chinchilla formula. These are the highest accuracy models for a compute budget and are available today open-source! (1/5)

Press: businesswire.com/news/home/2023… The AI industry is becoming increasingly closed. We believe in fostering open access to the most advanced models. Cerebras-GPT is being released under the Apache 2.0 license, allowing royalty-free use for research or commercial applications. (2/5)