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

Fix SDPA tests (#28552)

Browse Source 
* skip bf16 test if not supported by device

* fix

* fix bis

* use is_torch_bf16_available_on_device

* use is_torch_fp16_available_on_device

* fix & use public llama

* use 1b model

* fix flacky test

---------

Co-authored-by: Your Name <you@example.com>
This commit is contained in:
fxmarty
2024-01-17 17:29:18 +01:00
committed by GitHub
parent d6ffe74dfa
commit 2c1eebc121
2 changed files with 19 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/models/llama/test_modeling_llama.py
View File

@@ -457,10 +457,10 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
"""
max_new_tokens = 30
tokenizer = LlamaTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
tokenizer = LlamaTokenizer.from_pretrained("saibo/llama-1B")
model_sdpa = LlamaForCausalLM.from_pretrained(
"meta-llama/Llama-2-7b-hf",
"saibo/llama-1B",
torch_dtype=torch.float16,
low_cpu_mem_usage=True,
).to(torch_device)
@@ -468,7 +468,7 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
self.assertTrue(model_sdpa.config._attn_implementation == "sdpa")
model_eager = LlamaForCausalLM.from_pretrained(
"meta-llama/Llama-2-7b-hf",
"saibo/llama-1B",
torch_dtype=torch.float16,
low_cpu_mem_usage=True,
attn_implementation="eager",
@@ -488,7 +488,11 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
if not has_sdpa:
raise ValueError("The SDPA model should have SDPA attention layers")
texts = ["hi", "Hello this is a very long sentence my friend", "Today I am in Paris and"]
texts = [
"hi here's a longer context, getting longer and",
"Hello this is a very long sentence my friend, very long for real",
"Today I am in Paris and",
]
for padding_side in ["left", "right"]:
tokenizer.padding_side = padding_side