Mistral NeMo 12B: 128k context, Apache 2.0, drop-in replacement for Mistral 7B
In one sentence Mistral AI and NVIDIA release Mistral NeMo, a 12B model with up to 128k tokens of context, Apache 2.0 license, the new Tekken tokenizer and FP8 inference; designed as a drop-in replacement for Mistral 7B.
On July 18, 2024 Mistral AI and NVIDIA introduced Mistral NeMo, a 12-billion-parameter model built together. It has a context window of up to 128,000 tokens, and the weights (base and instruct) are released under the Apache 2.0 license.
Mistral describes it as a drop-in replacement for Mistral 7B: it uses a standard architecture, so anyone with a Mistral 7B-based system can swap it in without rewriting the pipeline.
NVIDIA also ships it as a NIM microservice and says it runs on a single L40S, GeForce RTX 4090 or RTX 4500 GPU.
Companies
Mistral AI, NVIDIA
Tools
Mistral NeMo 12B
Tags
Sources