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

remove phi3 empty cache (#10997)

Browse source
This commit is contained in:
Yishuo Wang 2024-05-13 14:09:55 +08:00 committed by GitHub
parent 99255fe36e
commit 1b3c7a6928
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 0 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
python/llm/src/ipex_llm/transformers/convert.py
View file

@ -188,13 +188,6 @@ def is_linear_module(module):
return result, (in_features, out_features, mp_group) return result, (in_features, out_features, mp_group)
def empty_cache_post(module, input, output):
try:
torch.xpu.empty_cache()
except: # cpu
pass
def convert_gptq(module, awq=False, llm_awq=False, act_order=False): def convert_gptq(module, awq=False, llm_awq=False, act_order=False):
from ipex_llm.transformers.low_bit_linear import get_block_size from ipex_llm.transformers.low_bit_linear import get_block_size
Q4_1 = get_block_size("asym_int4") Q4_1 = get_block_size("asym_int4")
@ -1524,8 +1517,6 @@ def _optimize_post(model, lightweight_bmm=False):
from ipex_llm.transformers.models.phi3 import model_forward_wrapper from ipex_llm.transformers.models.phi3 import model_forward_wrapper
model_forward = model_forward_wrapper(module.Phi3Model.forward) model_forward = model_forward_wrapper(module.Phi3Model.forward)
convert_forward(model, module.Phi3Model, model_forward) convert_forward(model, module.Phi3Model, model_forward)
# Empty cache after the first attention to run long context.
model.model.layers[0].self_attn.register_forward_hook(empty_cache_post)
elif model.config.model_type == 'yuan': elif model.config.model_type == 'yuan':
modeling_module_name = model.__class__.__module__ modeling_module_name = model.__class__.__module__
module = importlib.import_module(modeling_module_name) module = importlib.import_module(modeling_module_name)