Skip to content

Fix DeepSeek V4 FlashMLA auto KV cache dtype #45091

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hmellor merged 4 commits into from
Sep 3, 2026
Merged

Fix DeepSeek V4 FlashMLA auto KV cache dtype #45091

Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d3789b4
Fix DeepSeek V4 FlashMLA auto KV cache dtype
huangxh1618 Jun 9, 2026
d6bdc8c
Merge branch 'main' into fix/dsv4-flashmla-auto-kv-cache
Yuzu23 Jun 10, 2026
ebdac81
Merge main into fix/dsv4-flashmla-auto-kv-cache
khluu Sep 3, 2026
900fee7
Merge branch 'main' into fix/dsv4-flashmla-auto-kv-cache
tlrmchlsmth Sep 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 9 additions & 4 deletions vllm/models/deepseek_v4/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,10 +102,15 @@ def _resolve_dsv4_kv_cache_dtype(
"""
if use_fp8_ds_mla_layout:
# fp8_ds_mla block format: UE8M0 block-scaled fp8 packed as uint8.
assert kv_cache_dtype.startswith("fp8"), (
f"DeepseekV4 fp8_ds_mla layout only supports fp8 kv-cache, "
f"got {kv_cache_dtype}"
)
if kv_cache_dtype == "auto":
kv_cache_dtype = "fp8"
if not kv_cache_dtype.startswith("fp8"):
raise ValueError(
"DeepseekV4 fp8_ds_mla layout only supports fp8 "
f"kv-cache, got {kv_cache_dtype}. Please set "
"`--kv-cache-dtype fp8` or select a backend that supports "
"bfloat16 KV cache."
)
if kv_cache_dtype != "fp8_ds_mla":
if cache_config is not None:
cache_config.cache_dtype = "fp8_ds_mla"
Expand Down
Loading