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

Update BLOOM parameter counts (#18531)

Browse Source 
* Update BLOOM parameter counts

* Update BLOOM parameter counts
This commit is contained in:
Niklas Muennighoff
2022-08-12 19:36:18 +02:00
committed by GitHub
parent 153d1361c7
commit 56ef0ba447
6 changed files with 39 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
tests/models/bloom/test_modeling_bloom.py
View File

@@ -379,27 +379,27 @@ class BloomModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase)
def test_simple_generation(self):
# This test is a bit flaky. For some GPU architectures, pytorch sets by default allow_fp16_reduced_precision_reduction = True and some operations
# do not give the same results under this configuration, especially torch.baddmm and torch.bmm. https://pytorch.org/docs/stable/notes/numerical_accuracy.html#fp16-on-mi200
# As we leave the default value (True) for allow_fp16_reduced_precision_reduction , the tests failed when running in half-precision with smaller models (350m)
# As we leave the default value (True) for allow_fp16_reduced_precision_reduction , the tests failed when running in half-precision with smaller models (560m)
# Please see: https://pytorch.org/docs/stable/notes/cuda.html#reduced-precision-reduction-in-fp16-gemms
# This discrepancy is observed only when using small models and seems to be stable for larger models.
# Our conclusion is that these operations are flaky for small inputs but seems to be stable for larger inputs (for the functions `baddmm` and `bmm`), and therefore for larger models.
# Here is a summary of an ablation study of our observations
# EXPECTED_OUTPUT = "I enjoy walking with my cute dog, and I love to watch the kids play. I am a very active person, and I am a very good listener. I am a very good person, and I am a very good person. I am a"
# 350m + allow_fp16_reduced_precision_reduction = False + torch.bmm ==> PASS
# 350m + allow_fp16_reduced_precision_reduction = False + torch.baddm ==> PASS
# 350m + allow_fp16_reduced_precision_reduction = True + torch.baddm ==> PASS
# 350m + allow_fp16_reduced_precision_reduction = True + torch.bmm ==> FAIL
# 560m + allow_fp16_reduced_precision_reduction = False + torch.bmm ==> PASS
# 560m + allow_fp16_reduced_precision_reduction = False + torch.baddm ==> PASS
# 560m + allow_fp16_reduced_precision_reduction = True + torch.baddm ==> PASS
# 560m + allow_fp16_reduced_precision_reduction = True + torch.bmm ==> FAIL
# EXPECTED_OUTPUT = "I enjoy walking with my cute dog, but I also enjoy hiking, biking, and swimming. I love to cook and bake. I love to cook and bake. I love to cook and bake. I love to cook and bake. I love"
# >=760m + allow_fp16_reduced_precision_reduction = True + torch.baddm ==> PASS (for use_cache=True and use_cache=False)
# >=760m + allow_fp16_reduced_precision_reduction = True + torch.bmm ==> PASS
# >=760m + allow_fp16_reduced_precision_reduction = False + torch.bmm ==> PASS
# >=1b1 + allow_fp16_reduced_precision_reduction = True + torch.baddm ==> PASS (for use_cache=True and use_cache=False)
# >=1b1 + allow_fp16_reduced_precision_reduction = True + torch.bmm ==> PASS
# >=1b1 + allow_fp16_reduced_precision_reduction = False + torch.bmm ==> PASS
path_350m = "bigscience/bloom-350m"
model = BloomForCausalLM.from_pretrained(path_350m, use_cache=True, revision="gs555750").cuda()
path_560m = "bigscience/bloom-560m"
model = BloomForCausalLM.from_pretrained(path_560m, use_cache=True, revision="gs555750").cuda()
model = model.eval()
tokenizer = BloomTokenizerFast.from_pretrained(path_350m)
tokenizer = BloomTokenizerFast.from_pretrained(path_560m)
input_sentence = "I enjoy walking with my cute dog"
# This output has been obtained using fp32 model on the huggingface DGX workstation - NVIDIA A100 GPU
@@ -416,10 +416,10 @@ class BloomModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase)
@slow
@require_torch_gpu
def test_batch_generation(self):
path_350m = "bigscience/bloom-350m"
model = BloomForCausalLM.from_pretrained(path_350m, use_cache=True, revision="gs555750").cuda()
path_560m = "bigscience/bloom-560m"
model = BloomForCausalLM.from_pretrained(path_560m, use_cache=True, revision="gs555750").cuda()
model = model.eval()
tokenizer = BloomTokenizerFast.from_pretrained(path_350m, padding_side="left")
tokenizer = BloomTokenizerFast.from_pretrained(path_560m, padding_side="left")
input_sentence = ["I enjoy walking with my cute dog", "I enjoy walking with my cute dog"]
@@ -437,10 +437,10 @@ class BloomModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase)
@require_torch_gpu
def test_batch_generation_padd(self):
path_350m = "bigscience/bloom-350m"
model = BloomForCausalLM.from_pretrained(path_350m, use_cache=True, revision="gs555750").cuda()
path_560m = "bigscience/bloom-560m"
model = BloomForCausalLM.from_pretrained(path_560m, use_cache=True, revision="gs555750").cuda()
model = model.eval()
tokenizer = BloomTokenizerFast.from_pretrained(path_350m, padding_side="left")
tokenizer = BloomTokenizerFast.from_pretrained(path_560m, padding_side="left")
input_sentence = ["I enjoy walking with my cute dog", "Hello my name is"]
input_sentence_without_pad = "Hello my name is"