Mistral nemo was your only good model, nobody uses nemotron models because they are way too dumb for any real tasks, just reuse whatever methodology you used for nemo please. I own 20 NVIDIA shares.
Disagree, I use parakeet and nemo-nano-codec-22khz-0.6kbps-12.5fps daily. Parakeet is the fastest (english) ASR, and nemo-nano-codec is the best descrete audio codec model.
That being said, +1 I would love another Mistral-Nemo-like model. It's the most "human-like" chatbot model out there. It's main weakness is the small "usable" context window.
All that needs to be done is increase the parameter count, increase the pretraining context length, then run the job again with Mistral-Nemo's original pretrain and instruct datasets.
While I don't agree that "Mistral nemo was your only good model" (nemotron models are pretty good for technical production level tasks), Mistral Nemo is a standout when it comes to the balance between creativity, sounding "human" (natural), while still being rather capable for the size, even to this day. It's a great generalist chatbot that people don't dislike talking to. Nemotron models by and large have that grating synthetic assistant tone that everybody hates.
Guest
>Mistral nemo was your only good model
Disagree, I use parakeet and nemo-nano-codec-22khz-0.6kbps-12.5fps daily. Parakeet is the fastest (english) ASR, and nemo-nano-codec is the best descrete audio codec model.
That being said, +1 I would love another Mistral-Nemo-like model. It's the most "human-like" chatbot model out there. It's main weakness is the small "usable" context window.
+rep
also qat if possible the q1 of deepseek R1 when it came out was very popular for those without a server motherboard
All that needs to be done is increase the parameter count, increase the pretraining context length, then run the job again with Mistral-Nemo's original pretrain and instruct datasets.
The House of Rothschild is willing to commit funds towards this project if necessary.
While I don't agree that "Mistral nemo was your only good model" (nemotron models are pretty good for technical production level tasks), Mistral Nemo is a standout when it comes to the balance between creativity, sounding "human" (natural), while still being rather capable for the size, even to this day.
It's a great generalist chatbot that people don't dislike talking to. Nemotron models by and large have that grating synthetic assistant tone that everybody hates.
So yeah, make Mistral Nemo but larger and better.
🚀