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/persimmon/test_modeling_persimmon.py
View File

@@ -44,6 +44,7 @@ if is_torch_available():
AutoTokenizer,
PersimmonForCausalLM,
PersimmonForSequenceClassification,
PersimmonForTokenClassification,
PersimmonModel,
)
from transformers.models.persimmon.modeling_persimmon import (
@@ -283,12 +284,15 @@ class PersimmonModelTester:
@require_torch
class PersimmonModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
all_model_classes = (
(PersimmonModel, PersimmonForCausalLM, PersimmonForSequenceClassification) if is_torch_available() else ()
(PersimmonModel, PersimmonForCausalLM, PersimmonForSequenceClassification, PersimmonForTokenClassification)
if is_torch_available()
else ()
)
pipeline_model_mapping = (
{
"feature-extraction": PersimmonModel,
"text-classification": PersimmonForSequenceClassification,
"token-classification": PersimmonForTokenClassification,
# TODO (ydshieh): check why these two fail. Fix them or skip them in a better way.
# "text-generation": PersimmonForCausalLM,
# "zero-shot": PersimmonForSequenceClassification,
@@ -365,6 +369,22 @@ class PersimmonModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTester
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->Persimmon,llama->persimmon
def test_persimmon_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 = PersimmonForTokenClassification(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),
)
@unittest.skip("Persimmon buffers include complex numbers, which breaks this test")
# Copied from tests.models.llama.test_modeling_llama.LlamaModelTest.test_save_load_fast_init_from_base
def test_save_load_fast_init_from_base(self):