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

Tests: move generate tests to the right mixin and delete redundant tests (#34464)

Browse Source 
* tmp commit

* tmp commit

* cull overwrites of deleted tests

* typo

* more specific docstring

* make fixup

* parameterize at the top?

* correction

* more deletions :D

* tmp commit

* for VLMs too

* fix _check_outputs

* test nit

* make fixup

* fix another flaky

* test_generate_from_inputs_embeds -- handle missing attention mask
This commit is contained in:
Joao Gante
2024-10-30 10:59:08 +00:00
committed by GitHub
parent 913330ca9f
commit 8a734ea2c3
46 changed files with 265 additions and 2348 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
tests/models/llama/test_modeling_llama.py
View File

@@ -26,7 +26,6 @@ from transformers import AutoTokenizer, LlamaConfig, StaticCache, is_torch_avail
from transformers.generation.configuration_utils import GenerationConfig
from transformers.testing_utils import (
backend_empty_cache,
require_bitsandbytes,
require_flash_attn,
require_read_token,
require_torch,
@@ -316,9 +315,6 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
# This is because we are hitting edge cases with the causal_mask buffer
model_split_percents = [0.5, 0.7, 0.8]
# used in `test_torch_compile`
_torch_compile_test_ckpt = "meta-llama/Llama-2-7b-hf"
# used in `test_torch_compile_for_training`
_torch_compile_train_cls = LlamaForCausalLM if is_torch_available() else None
@@ -585,43 +581,6 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
with self.assertRaises(KeyError):
config = _reinitialize_config(base_config, {"rope_scaling": {"rope_type": "linear"}}) # missing "factor"
@require_flash_attn
@require_torch_gpu
@require_bitsandbytes
@pytest.mark.flash_attn_test
@require_read_token
@slow
def test_flash_attn_2_generate_padding_right(self):
"""
Overwritting the common test as the test is flaky on tiny models
"""
model = LlamaForCausalLM.from_pretrained(
"meta-llama/Llama-2-7b-hf",
load_in_4bit=True,
device_map={"": 0},
)
tokenizer = LlamaTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
texts = ["hi", "Hello this is a very long sentence"]
tokenizer.padding_side = "right"
tokenizer.pad_token = tokenizer.eos_token
inputs = tokenizer(texts, return_tensors="pt", padding=True).to(0)
output_native = model.generate(**inputs, max_new_tokens=20, do_sample=False)
output_native = tokenizer.batch_decode(output_native)
model = LlamaForCausalLM.from_pretrained(
"meta-llama/Llama-2-7b-hf", load_in_4bit=True, device_map={"": 0}, attn_implementation="flash_attention_2"
)
output_fa_2 = model.generate(**inputs, max_new_tokens=20, do_sample=False)
output_fa_2 = tokenizer.batch_decode(output_fa_2)
self.assertListEqual(output_native, output_fa_2)
@require_flash_attn
@require_torch_gpu
@slow