This is what i have https://sabareesh.com/posts/llm-rig/ All You Need is 4x 4090 GPUs to Train Your Own Model
4 4090s is easily 8000$, nothing to scoff at IMO
Imagine some SLI 16 x 1080
Could you explain what is your use case for training 1B models? Learning or perhaps fine tuning?
Learning, prototype and then scale it in to cloud. Also can be used as inference engine to train another model if you are using model as a judge for RL.
4 4090s is easily 8000$, nothing to scoff at IMO
Imagine some SLI 16 x 1080
Could you explain what is your use case for training 1B models? Learning or perhaps fine tuning?
Learning, prototype and then scale it in to cloud. Also can be used as inference engine to train another model if you are using model as a judge for RL.