[V0.9.1] torchair_graph bugfix when chunked_prefill is true #1748
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
Signed-off-by: fems14 <1804143737@qq.com>
fems14
changed the title
| num_tokens, self.num_kv_heads, -1), | ||
| value=prefill_k_pe, | ||
| key_cache=kv_cache[0], | ||
| value_cache=kv_cache[1], | ||
| slot_indices=slots) | ||
|
|
||
| if kv_cache[0].numel( |
There was a problem hiding this comment.
Can you solve this problem for good, and remove all those redundant reshape_and_cache operation, merge all the code path into one. All the rope_k and nope_k are supposed to be reshape and cached into the kv_cache right?
There was a problem hiding this comment.
When the graph mode (without chunked prefill) is enabled, the reshapeAndCache path is not taken; instead, the fusion operator is used. The code for this part, including the entire file (mla_v1), has already been refactored by someone. It has been evaluated and is no longer on the current branch, but it will be merged into the main branch later.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
What this PR does / why we need it?
when torchair_graph and chunked_prefill are both true, save the decode kv_cache.Does this PR introduce any user-facing change?
How was this patch tested?