Skip to content

FP8 FA fixes #381

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
gshtras merged 2 commits into from
Jan 23, 2025
Merged

FP8 FA fixes #381

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
27a3ab1
FP8 FA fixes
ilia-cher Jan 22, 2025
c09f332
linter
ilia-cher Jan 22, 2025
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
5 changes: 2 additions & 3 deletions vllm/attention/backends/rocm_flash_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -681,9 +681,8 @@ def forward(
seq_lens,
make_attn_mask=False) # type: ignore
full_scales = (
1.0 / layer._q_scale.item(),
1.0 / layer._k_scale.item(), 1.0 /
layer._v_scale.item(), 1.0 / layer._prob_scale.item(),
layer._q_scale.item(), layer._k_scale.item(),
layer._v_scale.item(), layer._prob_scale.item(),
fp8_out_scale.item()) if (
fp8_out_scale and layer._q_scale
and layer._prob_scale
Expand Down
9 changes: 7 additions & 2 deletions vllm/attention/ops/triton_flash_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -390,6 +390,8 @@ def get_autotune_configs():

autotune_configs, autotune_keys = get_autotune_configs()

float8_info = torch.finfo(torch.float8_e4m3fnuz)


@triton.autotune(
configs=autotune_configs,
Expand Down Expand Up @@ -451,6 +453,8 @@ def attn_fwd(
BIAS_TYPE: tl.constexpr,
ENABLE_DROPOUT: tl.constexpr,
RETURN_ENCODED_SOFTMAX: tl.constexpr,
FP8_MIN: tl.constexpr = float8_info.min,
FP8_MAX: tl.constexpr = float8_info.max,
):
start_m = tl.program_id(0)
off_h_q = tl.program_id(1)
Expand Down Expand Up @@ -733,6 +737,7 @@ def attn_fwd(
causal_start_idx = seqlen_q - seqlen_k
if USE_FP8:
acc *= o_descale
acc = tl.clamp(acc, FP8_MIN, FP8_MAX)
acc = acc.to(Out.type.element_ty)
if IS_CAUSAL: # noqa: SIM102
if causal_start_idx > start_m_idx and causal_start_idx < end_m_idx:
Expand Down Expand Up @@ -832,9 +837,9 @@ def forward(

def check_and_convert(t, scale):
if t.dtype != float8:
finfo = torch.finfo(float8)
descale = 1.0 / scale
ts = (t * descale).clamp(min=finfo.min, max=finfo.max)
ts = (t * descale).clamp(min=float8_info.min,
max=float8_info.max)
return ts.to(float8)
else:
return t
Expand Down