4 ms·

LoRA conversely has different downsides. LoRA can be used in two ways: merged or unmerged. Unmerged (which is how it's trained) incurs a non-trivial computation

LoRA conversely has different downsides. LoRA can be used in two ways: merged or unmerged. Unmerged (which is how it's trained) incurs a non-trivial computation cost. Merged means you are modifying the model weights, which means you are stuck with that one model on that device (though, this usually applies for most implementations for the unmerged versions too). The benefit of prompt and prefix tuning (note: these are two separate methods) is that you can serve different soft-prompts and soft-prefixes efficiently with a single shared set of model weights.

eternalban 4y ago
https://ar5iv.labs.arxiv.org/html/2106.09685/assets/x1.png https://ar5iv.labs.arxiv.org/html/2106.09685/assets/x1.png > incurs a non-trivial computation cost The hit seems to be in energy/cpu not time since the W0 computation is in parallel with the BAx. (My assumption based on the latency claims in paper.) So an issue in edge deployments (battery life, etc.). > you are stuck with that one model on that device Upfront I have 0 clue on the actual numbers, but from a purely software architecture pov [in unmerged setup], having that W0 forward process once with n distinct BAx paths (for distinct fine tunings!) would address that, no? [p.s. say an application that takes as input A/V+Txt, runs that through an Ensemble LoRA (ELoRA™ /g) which each participant contributing its own BAx finetuing processing, sharing the single pre-trained W0.]
- arugulum 4y ago
  > My assumption based on the latency claims in paper. The latency claims are based on the merged version, where the modifications are merged into the model weights. Hence there is no latency cost, since the final model has the same shape as the original. > having that W0 forward process once with n distinct BAx paths (for distinct fine tunings!) would address that, no? The tl;dr is that that works, but is more expensive. Not ridiculously more expensive, but certainly more expensive that processing a few additional tokens with prefix/prompt tuning.
edwardjhu 4y ago
> Merged means you are modifying the model weights, which means you are stuck with that one model on that device (though, this usually applies for most implementations for the unmerged versions too). If one is careful with floating point issues, it's straightforward to unmerge the weights. W_0 = W_1 - BA Yes, prompt-based methods don't involve swapping weights.
- arugulum 4y ago
  Right, it's mathematically easy (again, up to floating point issues) to recover the weights as needed, but in terms of distribution/serving I'm guessing the plan is to have the original weights and carry around the LoRA weights and merge as necessary. (Also, I'm assuming you're the first author of LoRA.)
  - edwardjhu 4y ago
    Yes, the plan is to keep the original weights in VRAM and merge/unmerge LoRA weights on the fly. You can even cache a large library of LoRA ckpts in RAM. Yup, I am!

eternalban 4y ago

https://ar5iv.labs.arxiv.org/html/2106.09685/assets/x1.png https://ar5iv.labs.arxiv.org/html/2106.09685/assets/x1.png > incurs a non-trivial computation cost The hit seems to be in energy/cpu not time since the W0 computation is in parallel with the BAx. (My assumption based on the latency claims in paper.) So an issue in edge deployments (battery life, etc.). > you are stuck with that one model on that device Upfront I have 0 clue on the actual numbers, but from a purely software architecture pov [in unmerged setup], having that W0 forward process once with n distinct BAx paths (for distinct fine tunings!) would address that, no? [p.s. say an application that takes as input A/V+Txt, runs that through an Ensemble LoRA (ELoRA™ /g) which each participant contributing its own BAx finetuing processing, sharing the single pre-trained W0.]

arugulum 4y ago
> My assumption based on the latency claims in paper. The latency claims are based on the merged version, where the modifications are merged into the model weights. Hence there is no latency cost, since the final model has the same shape as the original. > having that W0 forward process once with n distinct BAx paths (for distinct fine tunings!) would address that, no? The tl;dr is that that works, but is more expensive. Not ridiculously more expensive, but certainly more expensive that processing a few additional tokens with prefix/prompt tuning.

edwardjhu 4y ago

> Merged means you are modifying the model weights, which means you are stuck with that one model on that device (though, this usually applies for most implementations for the unmerged versions too). If one is careful with floating point issues, it's straightforward to unmerge the weights. W_0 = W_1 - BA Yes, prompt-based methods don't involve swapping weights.

arugulum 4y ago
Right, it's mathematically easy (again, up to floating point issues) to recover the weights as needed, but in terms of distribution/serving I'm guessing the plan is to have the original weights and carry around the LoRA weights and merge as necessary. (Also, I'm assuming you're the first author of LoRA.)
- edwardjhu 4y ago
  Yes, the plan is to keep the original weights in VRAM and merge/unmerge LoRA weights on the fly. You can even cache a large library of LoRA ckpts in RAM. Yup, I am!