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

[Docs] Fix warnings in mkdocs build (continued) (#23743)

Browse Source 
Signed-off-by: Zerohertz <ohg3417@gmail.com>
Signed-off-by: Hyogeun Oh (오효근) <ohg3417@gmail.com>
This commit is contained in:
Hyogeun Oh (오효근)
2025-08-28 02:17:29 +09:00
committed by GitHub
parent dd58932280
commit 4e4d017b6f
26 changed files with 56 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/v1/attention/backends/cpu_attn.py
View File

@@ -491,7 +491,8 @@ class TorchSDPABackendImpl(AttentionImpl[TorchSDPAMetadata]):
query: shape = [num_tokens, num_heads * head_size]
key: shape = [num_tokens, num_kv_heads * head_size]
value: shape = [num_tokens, num_kv_heads * head_size]
kv_cache = [2, num_blocks, block_size * num_kv_heads * head_size]
kv_cache: shape =
[2, num_blocks, block_size * num_kv_heads * head_size]
NOTE: kv_cache will be an empty tensor with shape [0]
for profiling run.
attn_metadata: Metadata for attention.

3
vllm/v1/attention/backends/flash_attn.py
View File

@@ -438,7 +438,8 @@ class FlashAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

8
vllm/v1/attention/backends/flashinfer.py
View File

@@ -637,11 +637,9 @@ class FlashInferImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache: shape -
# NHD: [num_blocks, 2, block_size, num_kv_heads, head_size]
# HND: [num_blocks, 2, num_kv_heads, block_size, head_size]
kv_cache: KV cache tensor with different possible shapes:
- NHD: [num_blocks, 2, block_size, num_kv_heads, head_size]
- HND: [num_blocks, 2, num_kv_heads, block_size, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

3
vllm/v1/attention/backends/flex_attention.py
View File

@@ -689,7 +689,8 @@ class FlexAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

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

@@ -235,7 +235,8 @@ class PallasAttentionBackendImpl(AttentionImpl):
query: shape = [num_tokens, num_heads * head_size]
key: shape = [num_tokens, num_kv_heads * head_size]
value: shape = [num_tokens, num_kv_heads * head_size]
kv_cache = [num_blocks, block_size, num_kv_heads * 2, head_size]
kv_cache: shape =
[num_blocks, block_size, num_kv_heads * 2, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]
@@ -329,7 +330,7 @@ def write_to_kv_cache(
Args:
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [num_blocks, block_size, num_kv_heads * 2, head_size]
kv_cache: shape = [num_blocks, block_size, num_kv_heads * 2, head_size]
num_slices_per_kv_cache_update_block: int
"""
_, page_size, num_combined_kv_heads, head_size = kv_cache.shape

3
vllm/v1/attention/backends/rocm_aiter_fa.py
View File

@@ -429,7 +429,8 @@ class AiterFlashAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

3
vllm/v1/attention/backends/tree_attn.py
View File

@@ -362,7 +362,8 @@ class TreeAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

3
vllm/v1/attention/backends/triton_attn.py
View File

@@ -285,7 +285,8 @@ class TritonAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]

3
vllm/v1/attention/backends/xformers.py
View File

@@ -330,7 +330,8 @@ class XFormersAttentionImpl(AttentionImpl):
query: shape = [num_tokens, num_heads, head_size]
key: shape = [num_tokens, num_kv_heads, head_size]
value: shape = [num_tokens, num_kv_heads, head_size]
kv_cache = [2, num_blocks, block_size, num_kv_heads, head_size]
kv_cache: shape =
[2, num_blocks, block_size, num_kv_heads, head_size]
attn_metadata: Metadata for attention.
Returns:
shape = [num_tokens, num_heads * head_size]