SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Remove all traces of low_cpu_mem_usage (#38792)

Browse Source 
* remove it from all py files

* remove it from the doc

* remove it from examples

* style

* remove traces of _fast_init

* Update test_peft_integration.py

* CIs
This commit is contained in:
Cyril Vallez
2025-06-12 16:39:33 +02:00
committed by GitHub
parent 3542e0b844
commit 4b8ec667e9
76 changed files with 100 additions and 598 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/models/t5/test_modeling_t5.py
View File

@@ -1109,14 +1109,16 @@ class T5ModelFp16Tests(unittest.TestCase):
# Load using `accelerate` in bf16
model = T5ForConditionalGeneration.from_pretrained(
"google-t5/t5-small", torch_dtype=torch.bfloat16, low_cpu_mem_usage=True
"google-t5/t5-small",
torch_dtype=torch.bfloat16,
)
self.assertTrue(model.decoder.block[0].layer[2].DenseReluDense.wo.weight.dtype == torch.bfloat16)
self.assertTrue(model.decoder.block[0].layer[2].DenseReluDense.wi.weight.dtype == torch.bfloat16)
# Load without using `accelerate`
model = T5ForConditionalGeneration.from_pretrained(
"google-t5/t5-small", torch_dtype=torch.float16, low_cpu_mem_usage=True
"google-t5/t5-small",
torch_dtype=torch.float16,
)
self.assertTrue(model.decoder.block[0].layer[2].DenseReluDense.wo.weight.dtype == torch.float32)
self.assertTrue(model.decoder.block[0].layer[2].DenseReluDense.wi.weight.dtype == torch.float16)