Instruct Models Release Date?
The blogpost published on March 21, 2025 mentions that Instruct models are coming soon. When will they be released? I have used the Base models, and they do not perform very well on certain tasks. I would love to use Instruct models in my research since I expect them to perform better. Thanks.
I wish Nvidia would respond to these directly. But anyway to my knowledge their paper does not mention a 47B instruct model, but they do have one (a reasoning model) here: https://huggingface.co/nvidia/Nemotron-H-47B-Reasoning-128K
I wish Nvidia would respond to these directly. But anyway to my knowledge their paper does not mention a 47B instruct model, but they do have one (a reasoning model) here: https://huggingface.co/nvidia/Nemotron-H-47B-Reasoning-128K
Yep, thanks! Page 18 of their paper mentions "In this work, we chose to build Nemotron-H-8B-VLM and Nemotron-H-56B-VLM on Nemotron-H-8B-Instruct and Nemotron-H-56B-Base (since Nemotron-H-56B-Instruct was unavailable)." Based on this, I am assuming there does not exist a 56B or 47B Instruct model at all, but hopefully they will release the 8B-Instruct model soon. Also, not sure if you are familiar with their Hymba models -- these are awesome, but due to the low interest from the public they did not release more of these models. Hopefully showing interest here will encourage them to create and release a 56B or 47B Instruct model as well!
Thanks for pointing that out. I am most interested in new high-quality base models I can fine tune and open instruct datasets like Nvidia's HelpSteer family which are awesome. I am still using their Llama 3.1 Nemotron 70B model because I keep finding issues with newer models around that size and cannot find smaller models that can compete with it yet.