Non-determinism in GPT-4 is caused by Sparse MoE
152334h.github.io
Non-determinism in GPT-4 is caused by Sparse MoE
1–10 of 186 posts
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#2I feel like this introduces the potential for weird and hard-to-implement side channel attacks, if the sequences in a batch can affect the routing of others.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#3I feel like this introduces the potential for weird and hard-to-implement side channel attacks, if the sequences in a batch can affect the routing of others.
I think you’re right. Would be very hard to exploit I imagine though.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#4This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor.
I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely differently, and have sort of a synonym-phrase thing going where some of the key words are the same.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#5I feel like this introduces the potential for weird and hard-to-implement side channel attacks, if the sequences in a batch can affect the routing of others.
I think you’re right. Would be very hard to exploit I imagine though.
Hard like building a virtual machine in an image decoder? If there’s a way there’s a will.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#6Mixture of Experts
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#7This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor. I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely dif…
GPT4 web chat for two hours a day? I buy that. Using the API repeatedly for the same inputs, eg developing a program, and the non-determinism is hard to miss.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#8This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor. I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely dif…
Thanks. I'm really no expert (:P) on MoE research; I just noticed what was written in the Soft MoE paper and felt a need to check.
The non-deterministic outputs are really similar, yeah, if you check the gist examples I linked https://gist.github.com/152334H/047827ad3740627f4d37826c867a.... This part is at least no surprise, since the randomness should be bounded.
I suspect OpenAI will figure out some way to reduce the randomness at some point, though, given their public commitment to eventually adding logprobs back to ChatCompletions.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#9This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor. I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely dif…
GPT4 web chat for two hours a day? I buy that. Using the API repeatedly for the same inputs, eg developing a program, and the non-determinism is hard to miss.
I would imagine that most people use nonzero temperature, so they won't need to look for any explanation for non-determinism.
Re: Non-determinism in GPT-4 is caused by Sparse MoE
#10This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor. I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely dif…
> I've been adamantly against MoE for a set of reasons
Such as?