Skip to content
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

Jump to bottom

Added offloading support FP8 attention #1131

Merged
merged 5 commits into from
Sep 5, 2024
Merged

Added offloading support FP8 attention #1131

Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
0ef5803
Added offloading support FP8 attention
Aug 23, 2024
84857bb
Merge branch 'main' into main
ksivaman Sep 4, 2024
e59e4ba
Update transformer_engine/pytorch/attention.py
sanandaraj5597 Sep 4, 2024
e54c03b
Fix
ksivaman Sep 5, 2024
4e249fb
Merge branch 'main' into main
ksivaman Sep 5, 2024
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
9 changes: 7 additions & 2 deletions transformer_engine/pytorch/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5637,16 +5637,21 @@ def forward(
out_save = out_ret
fp8_tensors = (None, None, None, None, None, None)

ctx.fp8 = fp8 and int(os.getenv("NVTE_FP8_DPA_BWD", "1"))

from .cpu_offload import CPUOffloadEnabled

if CPUOffloadEnabled:
tensor_list = [q, k, v, out_save, cu_seqlens_q, cu_seqlens_kv]
if ctx.fp8:
tensor_list = fp8_tensors.extend(aux_ctx_tensors)
else:
tensor_list = [q, k, v, out_save].extend(aux_ctx_tensors)

qkv_layout = "sbhd_sbhd_sbhd"
for tensor in tensor_list:
if tensor is not None:
tensor.activation_offloading = True

ctx.fp8 = fp8 and int(os.getenv("NVTE_FP8_DPA_BWD", "1"))
qkvo_tensors = (q, k, v, out_save) if not ctx.fp8 else (None, None, None, None)
ctx.save_for_backward(
*qkvo_tensors,
Expand Down
Loading