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

fix: providing a tensor to cache_position in model.generate kwargs always crashes because of boolean test (#39300)

Browse Source 
* fix: cache_position: RuntimeError: Boolean value of Tensor with more than one value is ambiguous

* test cache_position

* move test

* propagate changes

---------

Co-authored-by: Masataro Asai <guicho2.71828@gmail.com>
This commit is contained in:
Joao Gante
2025-07-30 18:30:28 +01:00
committed by GitHub
parent 9b3203f47b
commit 4f93cc9174
6 changed files with 150 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/models/dia/test_modeling_dia.py
View File

@@ -33,7 +33,7 @@ from transformers.testing_utils import (
from transformers.utils import is_soundfile_available, is_torch_available, is_torchaudio_available
from transformers.utils.import_utils import is_datasets_available
from ...generation.test_utils import GenerationTesterMixin
from ...generation.test_utils import GenerationTesterMixin, has_similar_generate_outputs
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
@@ -512,7 +512,7 @@ class DiaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
outputs_cached.scores = full_cached_scores
# The two sets of generated text and past kv should be equal to each other
self._check_similar_generate_outputs(outputs, outputs_cached)
self.assertTrue(has_similar_generate_outputs(outputs, outputs_cached))
for layer_idx in range(len(outputs_cached.past_key_values)):
for kv_idx in range(len(outputs_cached.past_key_values[layer_idx])):
self.assertTrue(

6
tests/models/kosmos2/test_modeling_kosmos2.py
View File

@@ -40,7 +40,7 @@ from transformers.utils import (
is_vision_available,
)
from ...generation.test_utils import GenerationTesterMixin
from ...generation.test_utils import GenerationTesterMixin, has_similar_generate_outputs
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import (
TEST_EAGER_MATCHES_SDPA_INFERENCE_PARAMETERIZATION,
@@ -650,7 +650,7 @@ class Kosmos2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
outputs_from_embeds = model.generate(
input_ids=input_ids, inputs_embeds=inputs_embeds, **generation_kwargs, **inputs_dict
)
self._check_similar_generate_outputs(outputs_from_ids, outputs_from_embeds)
self.assertTrue(has_similar_generate_outputs(outputs_from_ids, outputs_from_embeds))
# input_ids is not a required input on most models -- if we don't pass it, the newly generated tokens will
# be the same
@@ -658,7 +658,7 @@ class Kosmos2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
inputs_embeds=inputs_embeds, **generation_kwargs, **inputs_dict
)
outputs_from_embeds.sequences = outputs_from_embeds.sequences[:, inputs_embeds.shape[1] :]
self._check_similar_generate_outputs(outputs_from_embeds_wo_ids, outputs_from_embeds)
self.assertTrue(has_similar_generate_outputs(outputs_from_embeds_wo_ids, outputs_from_embeds))
# We will verify our results on an image of cute cats

6
tests/models/kyutai_speech_to_text/test_modeling_kyutai_speech_to_text.py
View File

@@ -38,7 +38,7 @@ from transformers.testing_utils import (
torch_device,
)
from ...generation.test_utils import GenerationTesterMixin
from ...generation.test_utils import GenerationTesterMixin, has_similar_generate_outputs
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import (
TEST_EAGER_MATCHES_SDPA_INFERENCE_PARAMETERIZATION,
@@ -527,7 +527,7 @@ class KyutaiSpeechToTextModelTest(ModelTesterMixin, GenerationTesterMixin, Pipel
outputs_cached.scores = full_cached_scores
# The two sets of generated text and past kv should be equal to each other
self._check_similar_generate_outputs(outputs, outputs_cached)
self.assertTrue(has_similar_generate_outputs(outputs, outputs_cached))
for layer_idx in range(len(outputs_cached.past_key_values)):
for kv_idx in range(len(outputs_cached.past_key_values[layer_idx])):
self.assertTrue(
@@ -613,7 +613,7 @@ class KyutaiSpeechToTextModelTest(ModelTesterMixin, GenerationTesterMixin, Pipel
del model_attn
gc.collect()
self._check_similar_generate_outputs(res_eager, res_attn, atol=1e-3, rtol=1e-3)
self.assertTrue(has_similar_generate_outputs(res_eager, res_attn, atol=1e-3, rtol=1e-3))
@require_torch

4
tests/models/t5gemma/test_modeling_t5gemma.py
View File

@@ -29,7 +29,7 @@ from transformers.testing_utils import (
torch_device,
)
from ...generation.test_utils import GenerationTesterMixin
from ...generation.test_utils import GenerationTesterMixin, has_similar_generate_outputs
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
@@ -1196,7 +1196,7 @@ class T5GemmaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
outputs_cached.scores = full_cached_scores
# The two sets of generated text and past kv should be equal to each other
self._check_similar_generate_outputs(outputs, outputs_cached)
self.assertTrue(has_similar_generate_outputs(outputs, outputs_cached))
for layer_idx in range(len(outputs_cached.past_key_values)):
for kv_idx in range(len(outputs_cached.past_key_values[layer_idx])):
self.assertTrue(