This repository was archived by the owner on Sep 10, 2025. It is now read-only.
[Distributed] Enable KV cache #1154
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchchat/1154
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New FailuresAs of commit 1e78f59 with merge base 16b3d64 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
Jack-Khuu
approved these changes
Sep 17, 2024
| self.max_batch_size = max_batch_size | ||
| for b in self.layers.values(): | ||
| b.attention.kv_cache = KVCache( | ||
| max_batch_size, max_seq_length, self.config.n_local_heads, head_dim |
There was a problem hiding this comment.
This is fine, just calling it out as a note to self
Local head_dim is defined as self.config.dim // self.config.n_heads.
In Attention, head_dim is pulled from config.head_dim. Recall that TransformerArgs (config) is also initialized as self.config.dim // self.config.n_heads
| for b in self.layers.values(): | ||
| b.attention.kv_cache = KVCache( | ||
| max_batch_size, max_seq_length, self.config.n_local_heads, head_dim | ||
| # Lower the setup_cache call to the attention module because tensor |
There was a problem hiding this comment.
head_dim as defined on 440 is no longer used, let's remove it
There was a problem hiding this comment.
Good point, thanks!
lessw2020
approved these changes
Sep 17, 2024
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously KV cache was turned off in distributed cases, enabling it now.
The cache size is related to TP degree -- because the heads are divided across ranks.
For this information to be available, we lowered KV cache instantiation from root level to attention layer (which actually owns the cache and knows how itself is being TP'lized).