biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[v1] Add Whisper model support (encoder-decoder) (#21088)

Browse Source 
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Co-authored-by: NickLucche <nlucches@redhat.com>
This commit is contained in:
Russell Bryant
2025-09-10 16:53:35 -04:00
committed by GitHub
parent 4db4426404
commit 37e8182bfe
31 changed files with 429 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/v1/attention/backends/rocm_aiter_fa.py
View File

@@ -236,11 +236,11 @@ class AiterFlashAttentionMetadataBuilder(
def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str],
vllm_config: VllmConfig, device: torch.device):
self.vllm_config = vllm_config
super().__init__(kv_cache_spec, layer_names, vllm_config, device)
self.model_config = vllm_config.model_config
self.parallel_config = vllm_config.parallel_config
self.cache_config = vllm_config.cache_config
self.device = device
self.num_heads_q = self.model_config.get_num_attention_heads(
self.parallel_config)
@@ -248,7 +248,6 @@ class AiterFlashAttentionMetadataBuilder(
self.parallel_config)
self.headdim = self.model_config.get_head_size()
self.block_size = kv_cache_spec.block_size
self.kv_cache_spec = kv_cache_spec
# Sliding window size to be used with the AOT scheduler will be
# populated on first build() call.
self.aot_sliding_window: Optional[tuple[int, int]] = None