ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

tiny fix

Browse Source 
Former-commit-id: c0e9c0484dae6db93cef5048bad827ff22b1986a
This commit is contained in:
hiyouga
2024-09-05 23:41:16 +08:00
parent c5ef52a67a
commit 27547355e6
5 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llamafactory/model/model_utils/longlora.py
View File

@@ -353,7 +353,7 @@ def llama_sdpa_attention_forward(
def _apply_llama_patch() -> None:
require_version("transformers>=4.41.2,<=4.44.3", "To fix: pip install transformers>=4.41.2,<=4.44.3")
require_version("transformers>=4.41.2,<=4.45.0", "To fix: pip install transformers>=4.41.2,<=4.45.0")
LlamaAttention.forward = llama_attention_forward
LlamaFlashAttention2.forward = llama_flash_attention_2_forward
LlamaSdpaAttention.forward = llama_sdpa_attention_forward

2
src/llamafactory/model/model_utils/packing.py
View File

@@ -114,7 +114,7 @@ def get_unpad_data(attention_mask: "torch.Tensor") -> Tuple["torch.Tensor", "tor
def _patch_for_block_diag_attn(model_type: str) -> None:
require_version("transformers>=4.41.2,<=4.44.3", "To fix: pip install transformers>=4.41.2,<=4.44.3")
require_version("transformers>=4.41.2,<=4.45.0", "To fix: pip install transformers>=4.41.2,<=4.45.0")
if is_transformers_version_greater_than_4_43():
import transformers.modeling_flash_attention_utils