github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-15 22:19:38 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Fix/Inference] Fix GQA Triton and Support Llama3 (#5624)

Browse Source 
* [fix] GQA calling of flash decoding triton

* fix kv cache alloc shape

* fix rotary triton - GQA

* fix sequence max length assigning

* Sequence max length logic

* fix scheduling and spec-dec

* skip without import error

* fix pytest - skip without ImportError

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
Yuanheng Zhao
2024-04-23 13:09:55 +08:00
committed by GitHub
parent ccf72797e3
commit 5d4c1fe8f5
9 changed files with 183 additions and 194 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
colossalai/inference/batch_bucket.py
View File

@@ -386,6 +386,7 @@ class BatchBucket:
seq_id, seq = next(seqs_iter)
assert seq.output_len >= n_tokens, "Revoking len exceeds the current output len of the sequence"
seq.output_token_id = seq.output_token_id[:-n_tokens]
seq.revoke_finished_status()
self._sequence_lengths[self._sequences_indexes[seq_id]] -= n_tokens
def clear(self, free_block_tables_fn: Optional[Callable[[torch.Tensor], None]]) -> List[int]: