mirror of
https://github.com/kvcache-ai/ktransformers.git
synced 2026-11-05 01:07:28 +00:00
[GH-ISSUE #47] Can I run llama3.1 70b with rtx4090+64g ddr5 ram? #26
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
kvcache-ai/ktransformers#26
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @codeMonkey-shin on GitHub (Aug 20, 2024).
Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/47
Can I run llama3.1 70b with rtx4090+64g ddr5 ram?
At what rate per second are tokens generated?
@ELigoP commented on GitHub (Aug 21, 2024):
This is not an issue, it is general question.
As far as I know, ktransformers speeds up only MoE models, and llama3.1 70b is dense model (no routing/expert layers).
The answer to your question depends on your model quantization level, motherboard, RAM used and CPU.
@Atream commented on GitHub (Feb 10, 2025):
KTransformers does not offer as significant a performance improvement over llama.cpp on dense model as MOE models do, but it still performs well. You can try writing a YAML file yourself to place the most computationally intensive tasks on the GPU, while the remaining tasks can be completed on the CPU. As long as the DRAM can fully accommodate your GGUF file, KTransformers will be able to run.