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

use flash_attn xentropy when available #525

Merged
merged 2 commits into from
Sep 4, 2023
Merged

use flash_attn xentropy when available #525

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cdeba07
use flash_attn xentropy when available
tmm1 Sep 4, 2023
abd4f9a
log when xentropy is not found
tmm1 Sep 4, 2023
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
15 changes: 15 additions & 0 deletions src/axolotl/monkeypatch/llama_attn_hijack_flash.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,9 @@

# copied from https://github.com/lm-sys/FastChat/blob/main/fastchat/train/llama_flash_attn_monkey_patch.py

import logging
import warnings
from functools import partial
from typing import List, Optional, Tuple, Union

import torch
Expand Down Expand Up @@ -33,6 +35,9 @@
)


LOG = logging.getLogger("axolotl")


def replace_llama_attn_with_flash_attn(packed: Optional[bool] = False):
transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = ( # pylint: disable=protected-access
_prepare_decoder_attention_mask
Expand All @@ -44,6 +49,16 @@ def replace_llama_attn_with_flash_attn(packed: Optional[bool] = False):
llama_model_forward
)

try:
from flash_attn.losses.cross_entropy import CrossEntropyLoss

LOG.info("patching with flash_attn.losses.cross_entropy")
transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
CrossEntropyLoss, inplace_backward=True
)
except ImportError:
pass
tmm1 marked this conversation as resolved.
Show resolved Hide resolved
tmm1 marked this conversation as resolved.
Show resolved Hide resolved


# Disable the transformation of the attention mask in LlamaModel as the flash attention
# requires the attention mask to be the same as the key_padding_mask
Expand Down