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

🚨 🚨 Inherited CausalLM Tests (#37590)

Browse Source 
* stash commit

* Experiment 1: Try just Gemma

* Experiment 1: Just try Gemma

* make fixup

* Trigger tests

* stash commit

* Try adding Gemma3 as well

* make fixup

* Correct attrib names

* Correct pipeline model mapping

* Add in all_model_classes for Gemma1 again

* Move the pipeline model mapping around again

* make fixup

* Revert Gemma3 changes since it's a VLM

* Let's try Falcon

* Correct attributes

* Correct attributes

* Let's try just overriding get_config() for now

* Do Nemotron too

* And Llama!

* Do llama/persimmon

* Correctly skip tests

* Fix Persimmon

* Include Phimoe

* Fix Gemma2

* Set model_tester_class correctly

* Add GLM

* More models!

* models models models

* make fixup

* Add Qwen3 + Qwen3MoE

* Correct import

* make fixup

* Add the QuestionAnswering classes

* Add the QuestionAnswering classes

* Move pipeline mapping to the right place

* Jetmoe too

* Stop RoPE testing models with no RoPE

* Fix up JetMOE a bit

* Fix up JetMOE a bit

* Can we just force pad_token_id all the time?

* make fixup

* fix starcoder2

* Move pipeline mapping

* Fix RoPE skipping

* Fix RecurrentGemma tests

* Fix Falcon tests

* Add MoE attributes

* Fix values for RoPE testing

* Make sure we set bos_token_id and eos_token_id in an appropriate range

* make fixup

* Fix GLM4

* Add mamba attributes

* Revert bits of JetMOE

* Re-add the JetMOE skips

* Update tests/causal_lm_tester.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Add licence

---------

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
This commit is contained in:
Matt
2025-05-23 18:29:31 +01:00
committed by GitHub
parent d5f992f5e6
commit 53fb245eb6
25 changed files with 816 additions and 4422 deletions
Download Patch File Download Diff File Expand all files Collapse all files

248
tests/models/glm/test_modeling_glm.py
View File

@@ -19,7 +19,6 @@ import pytest
from transformers import AutoModelForCausalLM, AutoTokenizer, GlmConfig, is_torch_available
from transformers.testing_utils import (
is_flaky,
require_flash_attn,
require_torch,
require_torch_large_accelerator,
@@ -28,10 +27,7 @@ from transformers.testing_utils import (
torch_device,
)
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
from ...causal_lm_tester import CausalLMModelTest, CausalLMModelTester
if is_torch_available():
@@ -46,133 +42,17 @@ if is_torch_available():
@require_torch
class GlmModelTester:
class GlmModelTester(CausalLMModelTester):
config_class = GlmConfig
if is_torch_available():
model_class = GlmModel
for_causal_lm_class = GlmForCausalLM
for_sequence_class = GlmForSequenceClassification
for_token_class = GlmForTokenClassification
def __init__(
self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=False,
use_labels=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=2,
num_attention_heads=4,
num_key_value_heads=2,
intermediate_size=37,
hidden_act="silu",
attention_dropout=0.1,
max_position_embeddings=512,
type_vocab_size=16,
type_sequence_label_size=2,
initializer_range=0.02,
num_labels=3,
num_choices=4,
pad_token_id=0,
scope=None,
):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.use_labels = use_labels
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.attention_dropout = attention_dropout
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.num_labels = num_labels
self.num_choices = num_choices
self.pad_token_id = pad_token_id
self.scope = scope
self.head_dim = self.hidden_size // self.num_attention_heads
# Copied from tests.models.mistral.test_modeling_mistral.MistralModelTester.prepare_config_and_inputs
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = torch.tril(torch.ones_like(input_ids).to(torch_device))
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
sequence_labels = None
token_labels = None
choice_labels = None
if self.use_labels:
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = self.get_config()
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
def get_config(self):
return self.config_class(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
num_key_value_heads=self.num_key_value_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
attention_dropout=self.attention_dropout,
max_position_embeddings=self.max_position_embeddings,
initializer_range=self.initializer_range,
pad_token_id=self.pad_token_id,
head_dim=self.head_dim,
)
def create_and_check_model(
self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
):
model = self.model_class(config=config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask)
result = model(input_ids)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
# Copied from tests.models.llama.test_modeling_llama.LlamaModelTester.prepare_config_and_inputs_for_common with Llama->Glm
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
token_type_ids,
input_mask,
sequence_labels,
token_labels,
choice_labels,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask}
return config, inputs_dict
base_model_class = GlmModel
causal_lm_class = GlmForCausalLM
sequence_class = GlmForSequenceClassification
token_class = GlmForTokenClassification
@require_torch
class GlmModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
class GlmModelTest(CausalLMModelTest, unittest.TestCase):
all_model_classes = (
(GlmModel, GlmForCausalLM, GlmForSequenceClassification, GlmForTokenClassification)
if is_torch_available()
@@ -188,120 +68,10 @@ class GlmModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
if is_torch_available()
else {}
)
test_headmasking = False
test_pruning = False
def setUp(self):
self.model_tester = GlmModelTester(self)
self.config_tester = ConfigTester(self, config_class=GlmConfig, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_model_various_embeddings(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
for type in ["absolute", "relative_key", "relative_key_query"]:
config_and_inputs[0].position_embedding_type = type
self.model_tester.create_and_check_model(*config_and_inputs)
def test_Glm_sequence_classification_model(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
print(config)
config.num_labels = 3
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor([self.model_tester.batch_size], self.model_tester.type_sequence_label_size)
model = self.model_tester.for_sequence_class(config)
model.to(torch_device)
model.eval()
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))
def test_Glm_sequence_classification_model_for_single_label(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
config.problem_type = "single_label_classification"
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor([self.model_tester.batch_size], self.model_tester.type_sequence_label_size)
model = self.model_tester.for_sequence_class(config)
model.to(torch_device)
model.eval()
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))
def test_Glm_sequence_classification_model_for_multi_label(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.num_labels = 3
config.problem_type = "multi_label_classification"
input_ids = input_dict["input_ids"]
attention_mask = input_ids.ne(1).to(torch_device)
sequence_labels = ids_tensor(
[self.model_tester.batch_size, config.num_labels], self.model_tester.type_sequence_label_size
).to(torch.float)
model = self.model_tester.for_sequence_class(config)
model.to(torch_device)
model.eval()
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))
def test_Glm_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 = self.model_tester.for_token_class(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),
)
@is_flaky()
def test_custom_4d_attention_mask(self):
"""Overwrite the common test to use atol=1e-3 instead of 1e-4. Can still rarely fail, thus flaky."""
for model_class in self.all_generative_model_classes:
if not model_class._supports_static_cache:
self.skipTest(f"{model_class.__name__} is not guaranteed to work with custom 4D attention masks")
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
if getattr(config, "sliding_window", 0) is not None and getattr(config, "sliding_window", 0) > 0:
self.skipTest(f"{model_class.__name__} with sliding window attention is not supported by this test")
model = model_class(config).to(device=torch_device, dtype=torch.float32)
(
input_ids,
position_ids,
input_ids_shared_prefix,
mask_shared_prefix,
position_ids_shared_prefix,
) = self._get_custom_4d_mask_test_data()
logits = model.forward(input_ids, position_ids=position_ids).logits
# logits.shape == torch.Size([3, 4, ...])
logits_shared_prefix = model(
input_ids_shared_prefix,
attention_mask=mask_shared_prefix,
position_ids=position_ids_shared_prefix,
)[0]
# logits_shared_prefix.shape == torch.Size([1, 6, ...])
out_last_tokens = logits[:, -1, :] # last tokens in each batch line
out_shared_prefix_last_tokens = logits_shared_prefix[0, -3:, :] # last three tokens
# comparing softmax-normalized logits:
normalized_0 = torch.nn.functional.softmax(out_last_tokens)
normalized_1 = torch.nn.functional.softmax(out_shared_prefix_last_tokens)
print(torch.abs(normalized_0 - normalized_1).max())
torch.testing.assert_close(normalized_0, normalized_1, rtol=1e-3, atol=1e-3)
model_tester_class = GlmModelTester
@slow