3 ms·

Very interesting. The state management is the really insightful find here. I always wondered how these large AI companies managed access for millions of simult

by glitchc 4mo ago

Very interesting. The state management is the really insightful find here. I always wondered how these large AI companies managed access for millions of simultaneous users without having to allocate a dedicated LLM instance for each user. Pushing the complete state down to the user after every call makes perfect sense. The LLM itself stays memoryless and ready to respond to an arbitrary prompt. Very nice.

geocar 4mo ago
N.B. This is exactly how seaside, vba, and even arc[1] do server-side state generally: by encrypting the blob-representing-state and sending to the client to be sent back on future requests (where it will be decrypted and rehydrated). It's an old trick that everyone designing protocols should know, since there are lots of applications beyond AI companies. [1]: As in, pg's lisp: https://arclanguage.github.io/ref/srv.html#:~:text=The%20previous%20section https://arclanguage.github.io/ref/srv.html#:~:text=The%20pre...
- tn1 4mo ago
  And don't forget the venerable .NET Forms with its kilobytes of __VIEWSTATE
  - antonvs 4mo ago
    > kilobytes cute
- LoganDark 4mo ago
  Do they mitigate replay attacks?
- vachina 4mo ago
  Also basically jwt tokens
b65e8bee43c2ed0 4mo ago
the exchange rate between text and its representation in memory is brutal. here's a bit from a recent article: >An 82 GB footprint in DDR3 on a 2016 Xeon. About 25 GB of weights and 56 GB of KV cache at the full 262K context. The KV cache is larger than the model. 262k tokens is not much at all. with ~5 characters per token, that's only 1.3 MB of plaintext.
- londons_explore 4mo ago
  The providers must have a more efficient approach. Most cache every request for 12+ hours, and they certainly can't spare 100GB of ram per request for 12 hours.
  - choppaface 4mo ago
    or maybe they don’t actually cache (fully) but lie and just don’t charge the user right now. at least half the users, who are probably also using the most similar tokens / prompts, wouldn’t really know the difference in latency (or care)
    - londons_explore 4mo ago
      If it actually cost that much RAM, they would almost certainly add extra things to the API to manage cache lifetime. Ie. A 'please cache this for X minutes' flag, or a setting for a single re-use cache (the most common use case)
      - cyanydeez 4mo ago
        https://platform.claude.com/docs/en/build-with-claude/prompt-caching https://platform.claude.com/docs/en/build-with-claude/prompt... suggests the can cache outside the gpu.
  - dist-epoch 4mo ago
    This is one reason why price of SSDs also doubled, not just of RAM. > LMCache extends the KV Cache from the NVIDIA GPU's fast HBM (Tier 1) to larger, more cost-effective tiers like CPU RAM and local SSDs. https://cloud.google.com/blog/topics/developers-practitioners/boosting-llm-performance-with-tiered-kv-cache-on-google-kubernetes-engine/ https://cloud.google.com/blog/topics/developers-practitioner...
londons_explore 4mo ago
Except the providers also cache the parsing of the prompt (the KV cache), and that has substantial cost savings (easily an 80% saving on typical coding use cases). That caching is done server side and not passed to the client. Which in turn means they still need state management on the server side, although it perhaps doesn't need the same level of global replication and availability.
- cyanydeez 4mo ago
  from the march changes, it looked like they increased cache eviction rates on the VRAM at claude causing everyone to start burning tokens as they had to regen token state.
cyanydeez 4mo ago
they still have to cache the tokens. its not completely stateless. in theory, every conversation is replayed from the beginning. in practice, its only going to be economical to heavily cache the stable portions of the text as tokens inside the GPU one of the reasons the Cloud providers have such heavy prompts is because that can be cached for all users, but its essentially poisonong the state before you even start. alot of the variability appears related to changing the context rather than the model. models are expensive and the bean counters know fine tuning and context changes are cheaper. id guess the IPOs are essentially the SOTA EOL.
mycall 4mo ago
While it seems like a good idea, resending a growing context window is very inefficient and costly. Instance pinning would make a huge efficiency gains but also collapse LLM provider revenue. This is something open models could better solve.
- brookst 4mo ago
  Can you elaborate? How could it be more efficient and bad for revenue? Would it also be bad for profit?
  - bruce343434 4mo ago
    More efficient in terms of bandwidth (not) used. More costly because it has to be stored somewhere instead.
- mswphd 4mo ago
  even a max size context window is what, ~1M? iirc tokens are generally part of a vocab of size ~300k. Assume no compression before the encryption (no clue if this is true, but compressing text before encryption can leak info regarding the message, namely how compressible it is), that's \log2 300k ~ 18 bits per token, or ~2 bytes. So each "turn" would involve ~2MB extra in each direction. And again, this is assuming max context. seems plausibly fine

by glitchc 4mo ago

geocar 4mo ago

N.B. This is exactly how seaside, vba, and even arc[1] do server-side state generally: by encrypting the blob-representing-state and sending to the client to be sent back on future requests (where it will be decrypted and rehydrated). It's an old trick that everyone designing protocols should know, since there are lots of applications beyond AI companies. [1]: As in, pg's lisp: https://arclanguage.github.io/ref/srv.html#:~:text=The%20previous%20section https://arclanguage.github.io/ref/srv.html#:~:text=The%20pre...

tn1 4mo ago
And don't forget the venerable .NET Forms with its kilobytes of __VIEWSTATE
- antonvs 4mo ago
  > kilobytes cute
LoganDark 4mo ago
Do they mitigate replay attacks?
vachina 4mo ago
Also basically jwt tokens

b65e8bee43c2ed0 4mo ago

the exchange rate between text and its representation in memory is brutal. here's a bit from a recent article: >An 82 GB footprint in DDR3 on a 2016 Xeon. About 25 GB of weights and 56 GB of KV cache at the full 262K context. The KV cache is larger than the model. 262k tokens is not much at all. with ~5 characters per token, that's only 1.3 MB of plaintext.

londons_explore 4mo ago
The providers must have a more efficient approach. Most cache every request for 12+ hours, and they certainly can't spare 100GB of ram per request for 12 hours.
- choppaface 4mo ago
  or maybe they don’t actually cache (fully) but lie and just don’t charge the user right now. at least half the users, who are probably also using the most similar tokens / prompts, wouldn’t really know the difference in latency (or care)
  - londons_explore 4mo ago
    If it actually cost that much RAM, they would almost certainly add extra things to the API to manage cache lifetime. Ie. A 'please cache this for X minutes' flag, or a setting for a single re-use cache (the most common use case)
    - cyanydeez 4mo ago
      https://platform.claude.com/docs/en/build-with-claude/prompt-caching https://platform.claude.com/docs/en/build-with-claude/prompt... suggests the can cache outside the gpu.
- dist-epoch 4mo ago
  This is one reason why price of SSDs also doubled, not just of RAM. > LMCache extends the KV Cache from the NVIDIA GPU's fast HBM (Tier 1) to larger, more cost-effective tiers like CPU RAM and local SSDs. https://cloud.google.com/blog/topics/developers-practitioners/boosting-llm-performance-with-tiered-kv-cache-on-google-kubernetes-engine/ https://cloud.google.com/blog/topics/developers-practitioner...

londons_explore 4mo ago

Except the providers also cache the parsing of the prompt (the KV cache), and that has substantial cost savings (easily an 80% saving on typical coding use cases). That caching is done server side and not passed to the client. Which in turn means they still need state management on the server side, although it perhaps doesn't need the same level of global replication and availability.

cyanydeez 4mo ago
from the march changes, it looked like they increased cache eviction rates on the VRAM at claude causing everyone to start burning tokens as they had to regen token state.

cyanydeez 4mo ago

they still have to cache the tokens. its not completely stateless. in theory, every conversation is replayed from the beginning. in practice, its only going to be economical to heavily cache the stable portions of the text as tokens inside the GPU one of the reasons the Cloud providers have such heavy prompts is because that can be cached for all users, but its essentially poisonong the state before you even start. alot of the variability appears related to changing the context rather than the model. models are expensive and the bean counters know fine tuning and context changes are cheaper. id guess the IPOs are essentially the SOTA EOL.

mycall 4mo ago

While it seems like a good idea, resending a growing context window is very inefficient and costly. Instance pinning would make a huge efficiency gains but also collapse LLM provider revenue. This is something open models could better solve.

brookst 4mo ago
Can you elaborate? How could it be more efficient and bad for revenue? Would it also be bad for profit?
- bruce343434 4mo ago
  More efficient in terms of bandwidth (not) used. More costly because it has to be stored somewhere instead.
mswphd 4mo ago
even a max size context window is what, ~1M? iirc tokens are generally part of a vocab of size ~300k. Assume no compression before the encryption (no clue if this is true, but compressing text before encryption can leak info regarding the message, namely how compressible it is), that's \log2 300k ~ 18 bits per token, or ~2 bytes. So each "turn" would involve ~2MB extra in each direction. And again, this is assuming max context. seems plausibly fine