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

3
tests/models/bamba/test_modeling_bamba.py
View File

@@ -556,7 +556,6 @@ class BambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
tmpdirname,
torch_dtype=torch.float16,
attn_implementation="flash_attention_2",
low_cpu_mem_usage=True,
)
.to(torch_device)
.eval()
@@ -600,7 +599,7 @@ class BambaModelIntegrationTest(unittest.TestCase):
@classmethod
def setUpClass(cls):
model_id = "ibm-fms/Bamba-9B"
cls.model = BambaForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, low_cpu_mem_usage=True)
cls.model = BambaForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
cls.tokenizer = AutoTokenizer.from_pretrained(model_id)
# feels a bit forced to have to do this for the generation test