ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

fix baichuan (#11606)

Browse source
This commit is contained in:
Zhao Changmin 2024-07-18 09:43:36 +08:00 committed by GitHub
parent bfcdc35b04
commit e5c0058c0e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
python/llm/src/ipex_llm/transformers/npu_models/baichuan.py
View file

@ -37,6 +37,7 @@ from torch.nn import functional as F
import importlib import importlib
from typing import Optional, Tuple from typing import Optional, Tuple
from ipex_llm.transformers.npu_models.common import merge_linear from ipex_llm.transformers.npu_models.common import merge_linear
from ipex_llm.transformers.models.utils import update_past_key_value
def merge_mlp(module: torch.nn.Module): def merge_mlp(module: torch.nn.Module):
@ -85,10 +86,10 @@ def baichuan_attention_fwd(
cos, sin, position_ids) cos, sin, position_ids)
# [bsz, nh, t, hd] # [bsz, nh, t, hd]
if past_key_value is not None: key_states, value_states = update_past_key_value(
# reuse k, v, self_attention past_key_value, key_states, value_states,
key_states = torch.cat([past_key_value[0], key_states], dim=2) kv_seq_len, False, "cpu"
value_states = torch.cat([past_key_value[1], value_states], dim=2) )
past_key_value = (key_states, value_states) if use_cache else None past_key_value = (key_states, value_states) if use_cache else None