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

[PyTorch] Improve CP P2P efficiency #1208

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[PyTorch] Improve CP P2P efficiency #1208

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 6 additions & 6 deletions transformer_engine/pytorch/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1561,7 +1561,7 @@ def forward(
fused_attn_qkv_dtype = TE_DType[q.dtype]
fused_attn_backend = FusedAttnBackend["F16_arbitrary_seqlen"]

p2p_comm_buffers = [None for _ in range(cp_size)]
p2p_comm_buffers = [None, None]
if use_fused_attention and qkv_format in ["bshd", "sbhd"]:
p2p_comm_buffers[0] = torch.cat((k.unsqueeze(-3), v.unsqueeze(-3)), dim=-3)
else:
Expand All @@ -1576,12 +1576,12 @@ def forward(
req.wait()

if i < (cp_size - 1):
p2p_comm_buffers[i + 1] = torch.empty_like(p2p_comm_buffers[i])
p2p_comm_buffers[(i + 1) % 2] = torch.empty_like(p2p_comm_buffers[i % 2])
send_recv_reqs[i % 2] = flash_attn_p2p_communicate(
rank,
p2p_comm_buffers[i],
p2p_comm_buffers[i % 2],
send_dst,
p2p_comm_buffers[i + 1],
p2p_comm_buffers[(i + 1) % 2],
recv_src,
cp_group,
batch_p2p_comm,
Expand All @@ -1592,11 +1592,11 @@ def forward(
or fp8_meta["recipe"].fp8_mha
or int(os.getenv("NVTE_FP8_DPA_BWD", "1"))
):
kv_inputs[i % 2] = p2p_comm_buffers[i]
kv_inputs[i % 2] = p2p_comm_buffers[i % 2]
else:
# KV exchange is in BF16/FP16, cast received KV in each step
kv_inputs[i % 2] = cast_to_fp8(
p2p_comm_buffers[i],
p2p_comm_buffers[i % 2],
fp8_meta["scaling_fwd"],
META_QKV,
fp8_dtype_forward,
Expand Down