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

Add TokenClassification for Mistral, Mixtral and Qwen2 (#29878)

Browse Source 
* Add MistralForTokenClassification

* Add tests and docs

* Add token classification for Mixtral and Qwen2

* Save llma for token classification draft

* Add token classification support for Llama, Gemma, Persimmon, StableLm and StarCoder2

* Formatting

* Add token classification support for Qwen2Moe model

* Add dropout layer to each ForTokenClassification model

* Add copied from in tests

* Update src/transformers/models/llama/modeling_llama.py

Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>

* Propagate suggested changes

* Style

---------

Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
This commit is contained in:
Joseph Enguehard
2024-05-20 09:06:57 +01:00
committed by GitHub
parent 481a957814
commit 07bf2dff78
39 changed files with 1174 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/models/stablelm/test_modeling_stablelm.py
View File

@@ -43,6 +43,7 @@ if is_torch_available():
AutoTokenizer,
StableLmForCausalLM,
StableLmForSequenceClassification,
StableLmForTokenClassification,
StableLmModel,
)
from transformers.models.stablelm.modeling_stablelm import (
@@ -287,12 +288,15 @@ class StableLmModelTester:
# Copied from transformers.tests.persimmon.test_modeling_persimmon.PersimmonModelTest with Persimmon -> StableLm
class StableLmModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
all_model_classes = (
(StableLmModel, StableLmForCausalLM, StableLmForSequenceClassification) if is_torch_available() else ()
(StableLmModel, StableLmForCausalLM, StableLmForSequenceClassification, StableLmForTokenClassification)
if is_torch_available()
else ()
)
pipeline_model_mapping = (
{
"feature-extraction": StableLmModel,
"text-classification": StableLmForSequenceClassification,
"token-classification": StableLmForTokenClassification,
# TODO (ydshieh): check why these two fail. Fix them or skip them in a better way.
# "text-generation": StableLmForCausalLM,
# "zero-shot": StableLmForSequenceClassification,
@@ -356,6 +360,22 @@ class StableLmModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
result = model(input_ids, attention_mask=attention_mask, labels=sequence_labels)
self.assertEqual(result.logits.shape, (self.model_tester.batch_size, self.model_tester.num_labels))
# Copied from tests.models.llama.test_modeling_llama.LlamaModelTest.test_llama_token_classification_model with Llama->StableLm,llama->stablelm
def test_stablelm_token_classification_model(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
token_labels = ids_tensor([self.model_tester.batch_size, self.model_tester.seq_length], config.num_labels)
model = StableLmForTokenClassification(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=attention_mask, labels=token_labels)
self.assertEqual(
result.logits.shape,
(self.model_tester.batch_size, self.model_tester.seq_length, self.model_tester.num_labels),
)
@parameterized.expand([("linear",), ("dynamic",)])
# Copied from tests.models.llama.test_modeling_llama.LlamaModelTest.test_model_rope_scaling_from_config with Llama->StableLm
def test_model_rope_scaling_from_config(self, scaling_type):