423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-15 03:10:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

update requires

Browse Source 
Former-commit-id: 77666bd227
This commit is contained in:
hiyouga
2024-10-29 16:10:07 +08:00
parent 6989b8c341
commit 163cf2ba5c
8 changed files with 50 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/llamafactory/model/model_utils/longlora.py
View File

@@ -86,7 +86,7 @@ def llama_attention_forward(
if getattr(self.config, "group_size_ratio", None) and self.training: # shift
groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
num_groups = q_len // groupsz
def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -195,7 +195,7 @@ def llama_flash_attention_2_forward(
if getattr(self.config, "group_size_ratio", None) and self.training: # shift
groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
num_groups = q_len // groupsz
def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -301,7 +301,7 @@ def llama_sdpa_attention_forward(
if getattr(self.config, "group_size_ratio", None) and self.training: # shift
groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
num_groups = q_len // groupsz
def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -353,7 +353,7 @@ def llama_sdpa_attention_forward(
def _apply_llama_patch() -> None:
require_version("transformers>=4.41.2,<=4.45.2", "To fix: pip install transformers>=4.41.2,<=4.45.2")
require_version("transformers>=4.41.2,<=4.46.0", "To fix: pip install transformers>=4.41.2,<=4.46.0")
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,<=4.45.2", "To fix: pip install transformers>=4.41.2,<=4.45.2")
require_version("transformers>=4.41.2,<=4.46.0", "To fix: pip install transformers>=4.41.2,<=4.46.0")
if is_transformers_version_greater_than_4_43():
import transformers.modeling_flash_attention_utils