Skip to content

[Sdpa / Flash] save the attention not a bool #28131

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

Closed
ArthurZucker wants to merge 1 commit into from
Closed

[Sdpa / Flash] save the attention not a bool #28131

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
7 changes: 3 additions & 4 deletions src/transformers/models/llama/modeling_llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -955,8 +955,7 @@ def __init__(self, config: LlamaConfig):
self.layers = nn.ModuleList(
[LlamaDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
)
self._use_sdpa = config._attn_implementation == "sdpa"
self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
self._attn_implementation = config._attn_implementation
self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

self.gradient_checkpointing = False
Expand Down Expand Up @@ -1024,10 +1023,10 @@ def forward(
if inputs_embeds is None:
inputs_embeds = self.embed_tokens(input_ids)

if self._use_flash_attention_2:
if self._attn_implementation == "flash_attention_2":
# 2d mask is passed through the layers
attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
elif self._use_sdpa and not output_attentions:
elif self._attn_implementation == "sdpa" and not output_attentions:
# output_attentions=True can not be supported when using SDPA, and we fall back on
# the manual implementation that requires a 4D causal mask in all cases.
attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
Expand Down