Update transformers tests generation util v4.45.2

conftest.py

@@ -1,3 +1,88 @@
+# coding=utf-8
+# Copyright 2020 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# tests directory-specific settings - this file is run automatically
+# by pytest before any tests are run
+import doctest
+import sys
+import warnings
+from os.path import abspath, dirname, join
+
+import _pytest
+import pytest
+from transformers.testing_utils import HfDoctestModule, HfDocTestParser
+
+
+NOT_DEVICE_TESTS = {
+    "test_tokenization",
+    "test_processor",
+    "test_processing",
+    "test_beam_constraints",
+    "test_configuration_utils",
+    "test_data_collator",
+    "test_trainer_callback",
+    "test_trainer_utils",
+    "test_feature_extraction",
+    "test_image_processing",
+    "test_image_processor",
+    "test_image_transforms",
+    "test_optimization",
+    "test_retrieval",
+    "test_config",
+    "test_from_pretrained_no_checkpoint",
+    "test_keep_in_fp32_modules",
+    "test_gradient_checkpointing_backward_compatibility",
+    "test_gradient_checkpointing_enable_disable",
+    "test_save_load_fast_init_from_base",
+    "test_fast_init_context_manager",
+    "test_fast_init_tied_embeddings",
+    "test_save_load_fast_init_to_base",
+    "test_torch_save_load",
+    "test_initialization",
+    "test_forward_signature",
+    "test_model_get_set_embeddings",
+    "test_model_main_input_name",
+    "test_correct_missing_keys",
+    "test_tie_model_weights",
+    "test_can_use_safetensors",
+    "test_load_save_without_tied_weights",
+    "test_tied_weights_keys",
+    "test_model_weights_reload_no_missing_tied_weights",
+    "test_pt_tf_model_equivalence",
+    "test_mismatched_shapes_have_properly_initialized_weights",
+    "test_matched_shapes_have_loaded_weights_when_some_mismatched_shapes_exist",
+    "test_model_is_small",
+    "test_tf_from_pt_safetensors",
+    "test_flax_from_pt_safetensors",
+    "ModelTest::test_pipeline_",  # None of the pipeline tests from PipelineTesterMixin (of which XxxModelTest inherits from) are running on device
+    "ModelTester::test_pipeline_",
+    "/repo_utils/",
+    "/utils/",
+    "/agents/",
+}
+
+# allow having multiple repository checkouts and not needing to remember to rerun
+# `pip install -e '.[dev]'` when switching between checkouts and running tests.
+git_repo_path = abspath(join(dirname(__file__), "src"))
+sys.path.insert(1, git_repo_path)
+
+# silence FutureWarning warnings in tests since often we can't act on them until
+# they become normal warnings - i.e. the tests still need to test the current functionality
+warnings.simplefilter(action="ignore", category=FutureWarning)
+
+
 class Secret:
     """
     Taken from: https://stackoverflow.com/a/67393351
@@ -13,8 +98,44 @@ def __str___(self):
         return "*******"
 
 
+def pytest_configure(config):
+    config.addinivalue_line(
+        "markers", "is_pt_tf_cross_test: mark test to run only when PT and TF interactions are tested"
+    )
+    config.addinivalue_line(
+        "markers", "is_pt_flax_cross_test: mark test to run only when PT and FLAX interactions are tested"
+    )
+    config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipelines are tested")
+    config.addinivalue_line("markers", "is_staging_test: mark test to run only in the staging environment")
+    config.addinivalue_line("markers", "accelerate_tests: mark test that require accelerate")
+    config.addinivalue_line("markers", "agent_tests: mark the agent tests that are run on their specific schedule")
+    config.addinivalue_line("markers", "not_device_test: mark the tests always running on cpu")
+
+
+def pytest_collection_modifyitems(items):
+    for item in items:
+        if any(test_name in item.nodeid for test_name in NOT_DEVICE_TESTS):
+            item.add_marker(pytest.mark.not_device_test)
+
+
 def pytest_addoption(parser):
-    parser.addoption("--token", action="store", default=None)
+    from transformers.testing_utils import pytest_addoption_shared
+
+    pytest_addoption_shared(parser)
+
+
+def pytest_terminal_summary(terminalreporter):
+    from transformers.testing_utils import pytest_terminal_summary_main
+
+    make_reports = terminalreporter.config.getoption("--make-reports")
+    if make_reports:
+        pytest_terminal_summary_main(terminalreporter, id=make_reports)
+
+
+def pytest_sessionfinish(session, exitstatus):
+    # If no tests are collected, pytest exists with code 5, which makes the CI fail.
+    if exitstatus == 5:
+        session.exitstatus = 0
 
 
 def pytest_generate_tests(metafunc):
@@ -23,3 +144,21 @@ def pytest_generate_tests(metafunc):
     option_value = Secret(metafunc.config.option.token)
     if "token" in metafunc.fixturenames:
         metafunc.parametrize("token", [option_value])
+
+
+# Doctest custom flag to ignore output.
+IGNORE_RESULT = doctest.register_optionflag("IGNORE_RESULT")
+
+OutputChecker = doctest.OutputChecker
+
+
+class CustomOutputChecker(OutputChecker):
+    def check_output(self, want, got, optionflags):
+        if IGNORE_RESULT & optionflags:
+            return True
+        return OutputChecker.check_output(self, want, got, optionflags)
+
+
+doctest.OutputChecker = CustomOutputChecker
+_pytest.doctest.DoctestModule = HfDoctestModule
+doctest.DocTestParser = HfDocTestParser

optimum/habana/transformers/generation/utils.py

@@ -199,19 +199,20 @@ def _prepare_decoder_input_ids_for_generation(
         # 2. `decoder_start_token_id` must have shape (batch_size, 1)
         if device is None:
             device = self.device
-        if token_idx is None:
-            if decoder_start_token_id.ndim == 1:
-                if decoder_start_token_id.shape[0] != batch_size:
-                    raise ValueError(
-                        f"`decoder_start_token_id` expected to have length {batch_size} but got {decoder_start_token_id.shape[0]}"
-                    )
-                decoder_start_token_id = decoder_start_token_id.view(-1, 1)
-            else:
-                decoder_start_token_id = (
-                    torch.ones((batch_size, 1), dtype=torch.long, device=device) * decoder_start_token_id
+        if decoder_start_token_id.ndim == 1:
+            if decoder_start_token_id.shape[0] != batch_size:
+                raise ValueError(
+                    f"`decoder_start_token_id` expected to have length {batch_size} but got {decoder_start_token_id.shape[0]}"
                 )
+            decoder_start_token_id = decoder_start_token_id.view(-1, 1)
         else:
-            # creating padded decoder_input_ids to achieve static shapes. Later new tokens once generated are copied in to decoder_input_ids based on token_idx
+            decoder_start_token_id = (
+                torch.ones((batch_size, 1), dtype=torch.long, device=device) * decoder_start_token_id
+            )
+
+        if token_idx is not None:
+            # creating padded decoder_input_ids to achieve static shapes.
+            # Later new tokens once generated are copied in to decoder_input_ids based on token_idx
             max_length = max_new_tokens + 1 if max_new_tokens is not None else self.generation_config.max_length
             decoder_start_token_id = (
                 torch.ones((batch_size, 1), dtype=torch.long, device=device) * decoder_start_token_id
@@ -2952,7 +2953,8 @@ def expand_if_needed(tensor, new_size, value, dim=-1):
                 if self.generation_config.early_stopping:
                     num_eos_tokens.add_(beam_tokens[0:num_beams].eq(self.config.eos_token_id).sum())
 
-                beam_scores.add_(torch.where(beam_tokens.eq(self.config.eos_token_id), float("-inf"), 0.0))
+                if self.config.eos_token_id is not None:
+                    beam_scores.add_(torch.where(beam_tokens.eq(self.config.eos_token_id), float("-inf"), 0.0))
                 beam_scores = beam_scores.view(batch_size, -1).unsqueeze(0)
                 _, selected = torch.topk(beam_scores, k=num_beams, dim=-1, largest=True, sorted=True)
                 offset = torch.arange(0, torch.numel(beam_scores), beam_scores.shape[-1]).unsqueeze(-1)
@@ -3089,6 +3091,9 @@ def move(obj, device):
             if not output_scores:
                 sequence_outputs["sequence_scores"] = None
 
+            if self.generation_config.static_shapes:
+                raise NotImplementedError("sequence_scores is not implemented for static_shapes")
+
             if self.config.is_encoder_decoder:
                 return GenerateBeamEncoderDecoderOutput(
                     sequences=sequence_outputs["sequences"],

optimum/habana/transformers/models/bart/modeling_bart.py

@@ -458,7 +458,9 @@ def gaudi_BartDecoder_forward(
 
     # past_key_values_length
     past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
-    tensor_past_key_values_length = token_idx - 1 if use_cache else torch.tensor(past_key_values_length)
+    tensor_past_key_values_length = (
+        token_idx - 1 if (use_cache and token_idx is not None) else torch.tensor(past_key_values_length)
+    )
 
     if inputs_embeds is None:
         inputs_embeds = self.embed_tokens(input)

pyproject.toml

@@ -41,3 +41,12 @@ skip-magic-trailing-comma = false
 
 # Like Black, automatically detect the appropriate line ending.
 line-ending = "auto"
+
+[tool.pytest.ini_options]
+addopts = "--doctest-glob='**/*.md'"
+doctest_optionflags="NUMBER NORMALIZE_WHITESPACE ELLIPSIS"
+markers = [
+    "flash_attn_test: marks tests related to flash attention (deselect with '-m \"not flash_attn_test\"')",
+    "bitsandbytes: select (or deselect with `not`) bitsandbytes integration tests",
+    "generate: marks tests that use the GenerationTesterMixin"
+]