From c7f62dbb8cb68cf9f1f0f5a5a24bc65ea1a76067 Mon Sep 17 00:00:00 2001 From: Cemberk Date: Thu, 6 Feb 2025 07:55:54 +0000 Subject: [PATCH] gfx1201 changes --- tests/generation/test_utils.py | 40 ++++++------- tests/models/aria/test_modeling_aria.py | 4 +- .../data2vec/test_modeling_data2vec_audio.py | 2 +- tests/models/dbrx/test_modeling_dbrx.py | 4 +- tests/models/dpr/test_modeling_dpr.py | 2 +- tests/models/emu3/test_modeling_emu3.py | 60 +++++++++---------- .../models/gpt_neox/test_modeling_gpt_neox.py | 4 +- .../granitemoe/test_modeling_granitemoe.py | 4 +- tests/models/idefics/test_modeling_idefics.py | 4 +- tests/models/llava/test_modeling_llava.py | 1 + tests/models/marian/test_modeling_marian.py | 2 + tests/models/moshi/test_modeling_moshi.py | 40 ++++++------- tests/models/olmo2/test_modeling_olmo2.py | 1 + tests/models/olmoe/test_modeling_olmoe.py | 4 +- tests/models/phimoe/test_modeling_phimoe.py | 2 +- .../qwen2_5_vl/test_modeling_qwen2_5_vl.py | 2 +- tests/models/roberta/test_modeling_roberta.py | 6 +- .../test_pipelines_audio_classification.py | 2 +- ..._pipelines_automatic_speech_recognition.py | 2 +- tests/pipelines/test_pipelines_common.py | 2 +- .../test_pipelines_depth_estimation.py | 2 +- .../pipelines/test_pipelines_image_to_text.py | 2 +- .../pipelines/test_pipelines_summarization.py | 2 +- .../test_pipelines_text_generation.py | 6 +- tests/tp/test_tp.py | 2 +- 25 files changed, 103 insertions(+), 99 deletions(-) diff --git a/tests/generation/test_utils.py b/tests/generation/test_utils.py index df3125cc4306..0b8ad3d61849 100644 --- a/tests/generation/test_utils.py +++ b/tests/generation/test_utils.py @@ -2546,30 +2546,30 @@ class GenerationIntegrationTests(unittest.TestCase, GenerationIntegrationTestsMi "return_tensors": "pt", } - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_custom_logits_processor(self): super().test_custom_logits_processor() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_max_new_tokens_encoder_decoder(self): super().test_max_new_tokens_encoder_decoder() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_eos_token_id_int_and_list_beam_search(self): super().test_eos_token_id_int_and_list_beam_search() pass - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101']) def test_transition_scores_greedy_search_normalized(self): super().test_transition_scores_greedy_search_normalized() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101']) def test_transition_scores_greedy_search(self): super().test_transition_scores_greedy_search() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_input_features_as_encoder_kwarg(self): super().test_generate_input_features_as_encoder_kwarg() @@ -2608,7 +2608,7 @@ def test_diverse_beam_search(self): ], ) - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_max_length_if_input_embeds(self): # PT-only test: TF doesn't have StoppingCriteria article = "Today a dragon flew over Paris." @@ -2623,7 +2623,7 @@ def test_max_length_if_input_embeds(self): out_gen_embeds = model.generate(inputs_embeds=inputs_embeds, max_length=max_length) self.assertEqual(out_gen.shape[-1], input_len + out_gen_embeds.shape[-1]) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_min_length_if_input_embeds(self): # PT-only test: TF doesn't have StoppingCriteria article = "Today a dragon flew over Paris." @@ -2676,7 +2676,7 @@ def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwa ) # TODO (joao): replace `stop_sequence` in the pipeline by the more recent `generate` functionality - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_stop_sequence_stopping_criteria(self): # PT-only test: TF doesn't have StoppingCriteria prompt = """Hello I believe in""" @@ -3212,7 +3212,7 @@ def test_decoder_start_id_from_config(self): with self.assertRaises(ValueError): outputs = bart_model.generate(input_ids, generation_config=GenerationConfig(do_sample=False)) - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101']) def test_contrastive_search_batched(self): # PT-only test: TF doesn't have constrained beam search # Tests that contrastive search works with batched inputs (i.e. has the same output as for non-batched inputs) @@ -3260,7 +3260,7 @@ def test_logits_processor_not_inplace(self): self.assertListEqual(out.logits[-1].tolist(), out.scores[-1].tolist()) self.assertNotEqual(out_with_temp.logits[-1].tolist(), out_with_temp.scores[-1].tolist()) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_eos_token_id_int_and_list_top_k_top_sampling(self): # Has TF equivalent: this test relies on random sampling generation_kwargs = { @@ -3289,7 +3289,7 @@ def test_eos_token_id_int_and_list_top_k_top_sampling(self): generated_tokens = model.generate(**tokens, eos_token_id=eos_token_id, **generation_kwargs) self.assertTrue(expectation == len(generated_tokens[0])) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_model_kwarg_encoder_signature_filtering(self): # Has TF equivalent: ample use of framework-specific code bart_tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-bart") @@ -3327,7 +3327,7 @@ def forward(self, input_ids, **kwargs): # FakeEncoder.forward() accepts **kwargs -> no filtering -> type error due to unexpected input "foo" bart_model.generate(input_ids, foo="bar") - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_default_max_length_warning(self): model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2").to(torch_device) tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2") @@ -3385,7 +3385,7 @@ def test_default_assisted_generation(self): self.assertEqual(config.assistant_confidence_threshold, 0.4) self.assertEqual(config.is_assistant, False) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_generated_length_assisted_generation(self): # PT-only test: TF doesn't support assisted decoding yet. model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2").to(torch_device) @@ -3414,7 +3414,7 @@ def test_generated_length_assisted_generation(self): ) self.assertTrue((input_length + 10) <= out.shape[-1] <= 20) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_model_kwarg_assisted_decoding_decoder_only(self): # PT-only test: TF doesn't support assisted decoding yet. model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2").to(torch_device) @@ -3449,7 +3449,7 @@ def test_model_kwarg_assisted_decoding_decoder_only(self): ) self.assertListEqual(outputs_assisted.tolist(), outputs_tti.tolist()) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_model_kwarg_assisted_decoding_encoder_decoder(self): """ Tests that the following scenario is compatible with assisted generation: @@ -3516,7 +3516,7 @@ def prepare_inputs_for_generation(self, *args, foo=False, encoder_outputs=None, ) self.assertListEqual(outputs_assisted.tolist(), outputs_foo.tolist()) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_assisted_decoding_encoder_decoder_shared_encoder(self): """ Tests that the following scenario is compatible with assisted generation: @@ -3595,7 +3595,7 @@ def prepare_inputs_for_generation(self, *args, foo=False, encoder_outputs=None, ) self.assertListEqual(outputs_assisted.tolist(), outputs_foo.tolist()) - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_assisted_decoding_num_assistant_tokens_heuristic_schedule(self): # This test ensures that the assisted generation num_assistant_tokens 'heuristic' schedule works properly. @@ -4093,7 +4093,7 @@ def test_padding_input_contrastive_search_t5(self): self.assertEqual(generated_text_no_padding, generated_text_with_padding) self.assertEqual(generated_text_no_padding, "Ich muss diese Aufgabe vor Ende des Tages beenden.") - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_prepare_inputs_for_generation_decoder_llm(self): """Tests GenerationMixin.prepare_inputs_for_generation against expected usage with decoder-only llms.""" @@ -4234,7 +4234,7 @@ def test_generate_compile_fullgraph_tiny(self): gen_out = compiled_generate(**model_inputs, generation_config=generation_config) self.assertTrue(gen_out.shape[1] > model_inputs["input_ids"].shape[1]) # some text was generated - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_assisted_generation_early_exit(self): """ Tests that assisted generation with early exit works as expected. Under the hood, this has complex cache diff --git a/tests/models/aria/test_modeling_aria.py b/tests/models/aria/test_modeling_aria.py index f02e3d0505bb..5452f7e5ef62 100644 --- a/tests/models/aria/test_modeling_aria.py +++ b/tests/models/aria/test_modeling_aria.py @@ -195,11 +195,11 @@ class AriaForConditionalGenerationModelTest(ModelTesterMixin, GenerationTesterMi test_head_masking = False _is_composite = True - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_flex_attention_with_grads(self): super().test_flex_attention_with_grads() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() diff --git a/tests/models/data2vec/test_modeling_data2vec_audio.py b/tests/models/data2vec/test_modeling_data2vec_audio.py index 6b422db46de4..a30e3bf15f20 100644 --- a/tests/models/data2vec/test_modeling_data2vec_audio.py +++ b/tests/models/data2vec/test_modeling_data2vec_audio.py @@ -22,7 +22,7 @@ from tests.test_modeling_common import floats_tensor, ids_tensor, random_attention_mask from transformers import Data2VecAudioConfig, is_torch_available -from transformers.testing_utils import is_pt_flax_cross_test, require_soundfile, require_torch, slow, torch_device +from transformers.testing_utils import is_pt_flax_cross_test, require_soundfile, require_torch, slow, torch_device, skipIfRocm from ...test_configuration_common import ConfigTester from ...test_modeling_common import ModelTesterMixin, _config_zero_init diff --git a/tests/models/dbrx/test_modeling_dbrx.py b/tests/models/dbrx/test_modeling_dbrx.py index 0900b5d75ca7..6a343dd3b0a4 100644 --- a/tests/models/dbrx/test_modeling_dbrx.py +++ b/tests/models/dbrx/test_modeling_dbrx.py @@ -327,12 +327,12 @@ class DbrxModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin test_headmasking = False test_pruning = False - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() pass - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() pass diff --git a/tests/models/dpr/test_modeling_dpr.py b/tests/models/dpr/test_modeling_dpr.py index 43675259b534..d1af45e779c3 100644 --- a/tests/models/dpr/test_modeling_dpr.py +++ b/tests/models/dpr/test_modeling_dpr.py @@ -191,7 +191,7 @@ class DPRModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase): test_pruning = False test_head_masking = False - @skipIfRocm(arch='gfx90a') + @skipIfRocm(arch=['gfx90a','gfx942']) def test_eager_matches_sdpa_inference_2_float32(self): super().test_eager_matches_sdpa_inference_2_float32() pass diff --git a/tests/models/emu3/test_modeling_emu3.py b/tests/models/emu3/test_modeling_emu3.py index 8fa40260c35c..596d884bdf9c 100644 --- a/tests/models/emu3/test_modeling_emu3.py +++ b/tests/models/emu3/test_modeling_emu3.py @@ -319,123 +319,123 @@ class Emu3Vision2TextModelTest(ModelTesterMixin, GenerationTesterMixin, Pipeline test_pruning = False fx_compatible = False - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_methods_with_logits_to_keep(self): super().test_generate_methods_with_logits_to_keep() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_attention_outputs(self): super().test_attention_outputs() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_batching_equivalence(self): super().test_batching_equivalence() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_beam_sample_generate(self): super().test_beam_sample_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_beam_sample_generate_dict_output(self): super().test_beam_sample_generate_dict_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_beam_search_generate(self): super().test_beam_search_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_beam_search_generate_dict_output(self): super().test_beam_search_generate_dict_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_beam_search_low_memory(self): super().test_beam_search_low_memory() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_constrained_beam_search_generate(self): super().test_constrained_beam_search_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_constrained_beam_search_generate_dict_output(self): super().test_constrained_beam_search_generate_dict_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_determinism(self): super().test_determinism() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_group_beam_search_generate(self): super().test_group_beam_search_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_group_beam_search_generate_dict_output(self): super().test_group_beam_search_generate_dict_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_hidden_states_output(self): super().test_hidden_states_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_model_outputs_equivalence(self): super().test_model_outputs_equivalence() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_offloaded_cache_implementation_0_offloaded(self): super().test_offloaded_cache_implementation_0_offloaded() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_resize_tokens_embeddings(self): super().test_resize_tokens_embeddings() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_retain_grad_hidden_states_attentions(self): super().test_retain_grad_hidden_states_attentions() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_feed_forward_chunking(self): super().test_feed_forward_chunking() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_flex_attention_with_grads(self): super().test_flex_attention_with_grads() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_forward_with_logits_to_keep(self): super().test_forward_with_logits_to_keep() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_greedy_generate(self): super().test_greedy_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_greedy_generate_dict_outputs(self): super().test_greedy_generate_dict_outputs() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_group_beam_search_generate(self): super().test_group_beam_search_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_sample_generate(self): super().test_sample_generate() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_sample_generate_dict_output(self): super().test_sample_generate_dict_output() - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_save_load(self): super().test_save_load() diff --git a/tests/models/gpt_neox/test_modeling_gpt_neox.py b/tests/models/gpt_neox/test_modeling_gpt_neox.py index fda06ee2a63b..1bae31451ca2 100644 --- a/tests/models/gpt_neox/test_modeling_gpt_neox.py +++ b/tests/models/gpt_neox/test_modeling_gpt_neox.py @@ -283,11 +283,11 @@ class GPTNeoXModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi test_model_parallel = False test_head_masking = False - @skipIfRocm(arch=['gfx90a','gfx942','gfx1100','gfx1200','gfx1201']) + @skipIfRocm(arch=['gfx90a','gfx942','gfx1100','gfx1101','gfx1200','gfx1201']) def test_flex_attention_with_grads(self): super().test_flex_attention_with_grads() - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() pass diff --git a/tests/models/granitemoe/test_modeling_granitemoe.py b/tests/models/granitemoe/test_modeling_granitemoe.py index 7c6312e1c9be..b9249d22167c 100644 --- a/tests/models/granitemoe/test_modeling_granitemoe.py +++ b/tests/models/granitemoe/test_modeling_granitemoe.py @@ -298,12 +298,12 @@ class GraniteMoeModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.Test # This is because we are hitting edge cases with the causal_mask buffer model_split_percents = [0.5, 0.7, 0.8] - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() pass diff --git a/tests/models/idefics/test_modeling_idefics.py b/tests/models/idefics/test_modeling_idefics.py index 4b90e73e330b..1346ba70c553 100644 --- a/tests/models/idefics/test_modeling_idefics.py +++ b/tests/models/idefics/test_modeling_idefics.py @@ -597,12 +597,12 @@ class IdeficsForVisionText2TextTest(IdeficsModelTest, GenerationTesterMixin, uni all_generative_model_classes = (IdeficsForVisionText2Text,) if is_torch_available() else () - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_generate_from_inputs_embeds_with_static_cache(self): super().test_generate_from_inputs_embeds_with_static_cache() pass - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() pass diff --git a/tests/models/llava/test_modeling_llava.py b/tests/models/llava/test_modeling_llava.py index 23663ee649a7..47d29233aa3b 100644 --- a/tests/models/llava/test_modeling_llava.py +++ b/tests/models/llava/test_modeling_llava.py @@ -507,6 +507,7 @@ def test_batched_generation(self): outputs = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False) self.assertEqual(outputs, EXPECTED_OUTPUT) + @skipIfRocm(arch='gfx90a') def test_tokenizer_integration(self): slow_tokenizer = AutoTokenizer.from_pretrained("liuhaotian/llava-v1.6-34b", use_fast=False) slow_tokenizer.add_tokens("", True) diff --git a/tests/models/marian/test_modeling_marian.py b/tests/models/marian/test_modeling_marian.py index aed5381fcc70..330e38173e01 100644 --- a/tests/models/marian/test_modeling_marian.py +++ b/tests/models/marian/test_modeling_marian.py @@ -27,6 +27,7 @@ require_torch_fp16, slow, torch_device, + skipIfRocm, ) from transformers.utils import cached_property @@ -511,6 +512,7 @@ def test_unk_support(self): expected = [t.unk_token_id, t.unk_token_id, t.eos_token_id] self.assertEqual(expected, ids) + @skipIfRocm(arch='gfx942') def test_pad_not_split(self): input_ids_w_pad = self.tokenizer(["I am a small frog "], return_tensors="pt").input_ids[0].tolist() expected_w_pad = [38, 121, 14, 697, 38848, self.tokenizer.pad_token_id, 0] # pad diff --git a/tests/models/moshi/test_modeling_moshi.py b/tests/models/moshi/test_modeling_moshi.py index 9c81b8732655..65f560a5da1d 100644 --- a/tests/models/moshi/test_modeling_moshi.py +++ b/tests/models/moshi/test_modeling_moshi.py @@ -532,102 +532,102 @@ class MoshiTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase): test_resize_embeddings = False test_torchscript = False - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_constrained_beam_search_generate_dict_output(self): super().test_constrained_beam_search_generate_dict_output() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_constrained_beam_search_generate(self): super().test_constrained_beam_search_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_search_low_memory(self): super().test_beam_search_low_memory() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_search_generate_dict_outputs_use_cache(self): super().test_beam_search_generate_dict_outputs_use_cache() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_search_generate_dict_output(self): super().test_beam_search_generate_dict_output() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_search_generate(self): super().test_beam_search_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_sample_generate_dict_output(self): super().test_beam_sample_generate_dict_output() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_beam_sample_generate(self): super().test_beam_sample_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_dola_decoding_sample(self): super().test_dola_decoding_sample() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_generate_from_inputs_embeds_0_greedy(self): super().test_generate_from_inputs_embeds_0_greedy() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_generate_from_inputs_embeds_1_beam_search(self): super().test_generate_from_inputs_embeds_1_beam_search() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_greedy_generate(self): super().test_greedy_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_greedy_generate_dict_outputs(self): super().test_greedy_generate_dict_outputs() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_greedy_generate_dict_outputs_use_cache(self): super().test_greedy_generate_dict_outputs_use_cache() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_group_beam_search_generate(self): super().test_group_beam_search_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_group_beam_search_generate_dict_output(self): super().test_group_beam_search_generate_dict_output() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_new_cache_format(self): super().test_new_cache_format_0() pass - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_offloaded_cache_implementation_0_offloaded(self): super().test_offloaded_cache_implementation_0_offloaded() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_sample_generate(self): super().test_sample_generate() pass - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_sample_generate_dict_output(self): super().test_sample_generate_dict_output() pass diff --git a/tests/models/olmo2/test_modeling_olmo2.py b/tests/models/olmo2/test_modeling_olmo2.py index bf5e86313d8a..6edf7970714a 100644 --- a/tests/models/olmo2/test_modeling_olmo2.py +++ b/tests/models/olmo2/test_modeling_olmo2.py @@ -27,6 +27,7 @@ require_torch, slow, torch_device, + skipIfRocm, ) from ...generation.test_utils import GenerationTesterMixin diff --git a/tests/models/olmoe/test_modeling_olmoe.py b/tests/models/olmoe/test_modeling_olmoe.py index 1a57d7bbaed7..3c111cbcc1b3 100644 --- a/tests/models/olmoe/test_modeling_olmoe.py +++ b/tests/models/olmoe/test_modeling_olmoe.py @@ -306,13 +306,13 @@ class OlmoeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi model_split_percents = [0.5, 0.7, 0.8] def test_generate_with_static_cache(self): - if rocmUtils.is_rocm_skippable(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']): + if rocmUtils.is_rocm_skippable(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']): torch._dynamo.config.capture_dynamic_output_shape_ops = True super().test_generate_with_static_cache() pass def test_generate_from_inputs_embeds_with_static_cache(self): - if rocmUtils.is_rocm_skippable(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']): + if rocmUtils.is_rocm_skippable(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']): torch._dynamo.config.capture_dynamic_output_shape_ops = True super().test_generate_from_inputs_embeds_with_static_cache() pass diff --git a/tests/models/phimoe/test_modeling_phimoe.py b/tests/models/phimoe/test_modeling_phimoe.py index e6b8aba70cc3..2e27ce61c183 100644 --- a/tests/models/phimoe/test_modeling_phimoe.py +++ b/tests/models/phimoe/test_modeling_phimoe.py @@ -354,7 +354,7 @@ class PhimoeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix test_headmasking = False test_pruning = False - @skipIfRocm(arch=['gfx90a','gfx942','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx90a','gfx942','gfx1100','gfx1101','gfx1201','gfx1200']) def test_generate_with_static_cache(self): super().test_generate_with_static_cache() diff --git a/tests/models/qwen2_5_vl/test_modeling_qwen2_5_vl.py b/tests/models/qwen2_5_vl/test_modeling_qwen2_5_vl.py index bc355e8eed92..d5572bdca171 100644 --- a/tests/models/qwen2_5_vl/test_modeling_qwen2_5_vl.py +++ b/tests/models/qwen2_5_vl/test_modeling_qwen2_5_vl.py @@ -245,7 +245,7 @@ def test_generate_with_static_cache(self): def test_generate_from_inputs_embeds_0_greedy(self): super().test_generate_from_inputs_embeds_0_greedy() - @skipIfRocm(arch=['gfx942','gfx90a']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1200','gfx1201']) def test_generate_from_inputs_embeds_1_beam_search(self): super().test_generate_from_inputs_embeds_1_beam_search() diff --git a/tests/models/roberta/test_modeling_roberta.py b/tests/models/roberta/test_modeling_roberta.py index 6a2d9494f6c5..819a12ee6d90 100644 --- a/tests/models/roberta/test_modeling_roberta.py +++ b/tests/models/roberta/test_modeling_roberta.py @@ -397,17 +397,17 @@ class RobertaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi fx_compatible = True model_split_percents = [0.5, 0.8, 0.9] - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_cpu_offload(self): super().test_cpu_offload() pass - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_disk_offload_bin(self): super().test_disk_offload_bin() pass - @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx90a','gfx942','gfx1100','gfx1101','gfx1200']) def test_disk_offload_safetensors(self): super().test_disk_offload_safetensors() pass diff --git a/tests/pipelines/test_pipelines_audio_classification.py b/tests/pipelines/test_pipelines_audio_classification.py index 2cac65b2c11a..fad92fa56c68 100644 --- a/tests/pipelines/test_pipelines_audio_classification.py +++ b/tests/pipelines/test_pipelines_audio_classification.py @@ -137,7 +137,7 @@ def test_small_model_pt(self): self.assertIn(nested_simplify(output, decimals=4), [EXPECTED_OUTPUT, EXPECTED_OUTPUT_PT_2]) @require_torch - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_small_model_pt_fp16(self): model = "anton-l/wav2vec2-random-tiny-classifier" diff --git a/tests/pipelines/test_pipelines_automatic_speech_recognition.py b/tests/pipelines/test_pipelines_automatic_speech_recognition.py index a8d964939332..b35ee9dd2473 100644 --- a/tests/pipelines/test_pipelines_automatic_speech_recognition.py +++ b/tests/pipelines/test_pipelines_automatic_speech_recognition.py @@ -239,7 +239,7 @@ def test_whisper_fp16(self): speech_recognizer(waveform) @require_torch - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_small_model_pt_seq2seq(self): speech_recognizer = pipeline( model="hf-internal-testing/tiny-random-speech-encoder-decoder", diff --git a/tests/pipelines/test_pipelines_common.py b/tests/pipelines/test_pipelines_common.py index 98436fbf5879..f53f9507c9c1 100644 --- a/tests/pipelines/test_pipelines_common.py +++ b/tests/pipelines/test_pipelines_common.py @@ -899,7 +899,7 @@ def new_forward(*args, **kwargs): self.assertEqual(self.COUNT, 1) @require_torch - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_custom_code_with_string_tokenizer(self): # This test checks for an edge case - tokenizer loading used to fail when using a custom code model # with a separate tokenizer that was passed as a repo name rather than a tokenizer object. diff --git a/tests/pipelines/test_pipelines_depth_estimation.py b/tests/pipelines/test_pipelines_depth_estimation.py index 93baee2b3d1b..39ae326c2942 100644 --- a/tests/pipelines/test_pipelines_depth_estimation.py +++ b/tests/pipelines/test_pipelines_depth_estimation.py @@ -139,7 +139,7 @@ def test_small_model_pt(self): self.skipTest(reason="There is not hf-internal-testing tiny model for either GLPN nor DPT") @require_torch - @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1201','gfx1200']) + @skipIfRocm(arch=['gfx942','gfx90a','gfx1100','gfx1101','gfx1201','gfx1200']) def test_multiprocess(self): depth_estimator = pipeline( model="hf-internal-testing/tiny-random-DepthAnythingForDepthEstimation", diff --git a/tests/pipelines/test_pipelines_image_to_text.py b/tests/pipelines/test_pipelines_image_to_text.py index 31145e71bbbc..866f3600d837 100644 --- a/tests/pipelines/test_pipelines_image_to_text.py +++ b/tests/pipelines/test_pipelines_image_to_text.py @@ -123,7 +123,7 @@ def test_small_model_tf(self): compare_pipeline_output_to_hub_spec(single_output, ImageToTextOutput) @require_torch - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_small_model_pt(self): pipe = pipeline("image-to-text", model="hf-internal-testing/tiny-random-vit-gpt2") image = "./tests/fixtures/tests_samples/COCO/000000039769.png" diff --git a/tests/pipelines/test_pipelines_summarization.py b/tests/pipelines/test_pipelines_summarization.py index 4d91bde303c0..c22eea9e3331 100644 --- a/tests/pipelines/test_pipelines_summarization.py +++ b/tests/pipelines/test_pipelines_summarization.py @@ -91,7 +91,7 @@ def run_pipeline_test(self, summarizer, _): outputs = summarizer("This " * 1000, truncation=TruncationStrategy.ONLY_FIRST) @require_torch - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_small_model_pt(self): summarizer = pipeline(task="summarization", model="sshleifer/tiny-mbart", framework="pt") outputs = summarizer("This is a small test") diff --git a/tests/pipelines/test_pipelines_text_generation.py b/tests/pipelines/test_pipelines_text_generation.py index 77138194d120..9a920c0336f1 100644 --- a/tests/pipelines/test_pipelines_text_generation.py +++ b/tests/pipelines/test_pipelines_text_generation.py @@ -43,7 +43,7 @@ class TextGenerationPipelineTests(unittest.TestCase): tf_model_mapping = TF_MODEL_FOR_CAUSAL_LM_MAPPING @require_torch - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_small_model_pt(self): text_generator = pipeline(task="text-generation", model="sshleifer/tiny-ctrl", framework="pt") # Using `do_sample=False` to force deterministic output @@ -441,7 +441,7 @@ def get_test_pipeline( ) return text_generator, ["This is a test", "Another test"] - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_stop_sequence_stopping_criteria(self): prompt = """Hello I believe in""" text_generator = pipeline("text-generation", model="hf-internal-testing/tiny-random-gpt2") @@ -556,7 +556,7 @@ def run_pipeline_test(self, text_generator, _): @require_torch @require_accelerate @require_torch_accelerator - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_small_model_pt_bloom_accelerate(self): import torch diff --git a/tests/tp/test_tp.py b/tests/tp/test_tp.py index 519b787882c1..4b5d0eec9511 100644 --- a/tests/tp/test_tp.py +++ b/tests/tp/test_tp.py @@ -32,7 +32,7 @@ class TestTensorParallel(TestCasePlus): @require_torch_multi_gpu - @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1200']) + @skipIfRocm(arch=['gfx1201','gfx942','gfx90a','gfx1100','gfx1101','gfx1200']) def test_tp(self): distributed_args = f"""--nproc_per_node={torch.cuda.device_count()} --master_port={get_torch_dist_unique_port()}