optimum-rbln 0.9.2rc0__tar.gz → 0.9.2rc1__tar.gz

optimum_rbln-0.9.2rc1/.github/version.yaml

@@ -0,0 +1 @@
+rebel_compiler_version: 0.9.2.dev201+g36bf7f17.prod

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: optimum-rbln
-Version: 0.9.2rc0
+Version: 0.9.2rc1
 Summary: Optimum RBLN is the interface between the HuggingFace Transformers and Diffusers libraries and RBLN accelerators. It provides a set of tools enabling easy model loading and inference on single and multiple rbln device settings for different downstream tasks.
 Project-URL: Homepage, https://rebellions.ai
 Project-URL: Documentation, https://docs.rbln.ai

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/src/optimum/rbln/__version__.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.9.2rc0'
-__version_tuple__ = version_tuple = (0, 9, 2, 'rc0')
+__version__ = version = '0.9.2rc1'
+__version_tuple__ = version_tuple = (0, 9, 2, 'rc1')
 
 __commit_id__ = commit_id = None

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/src/optimum/rbln/transformers/models/colpali/configuration_colpali.py

@@ -28,23 +28,23 @@ class RBLNColPaliForRetrievalConfig(RBLNModelConfig):
     including vision tower settings and multi-sequence length support.
 
     Example usage:
-    ```python
-    from optimum.rbln import RBLNColPaliForRetrieval, RBLNColPaliForRetrievalConfig
-
-    # Create a configuration object
-    config = RBLNColPaliForRetrievalConfig(
-        max_seq_lens=1152,
-        output_hidden_states=False,
-        tensor_parallel_size=4
-    )
-
-    # Use the configuration with from_pretrained
-    model = RBLNColPaliForRetrieval.from_pretrained(
-        "vidore/colpali-v1.3-hf",
-        export=True,
-        rbln_config=config
-    )
-    ```
+        ```python
+        from optimum.rbln import RBLNColPaliForRetrieval, RBLNColPaliForRetrievalConfig
+
+        # Create a configuration object
+        config = RBLNColPaliForRetrievalConfig(
+            max_seq_lens=1152,
+            output_hidden_states=False,
+            tensor_parallel_size=4
+        )
+
+        # Use the configuration with from_pretrained
+        model = RBLNColPaliForRetrieval.from_pretrained(
+            "vidore/colpali-v1.3-hf",
+            export=True,
+            rbln_config=config
+        )
+        ```
     """
 
     submodules = ["vision_tower"]

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/src/optimum/rbln/transformers/models/colqwen2/configuration_colqwen2.py

@@ -20,6 +20,37 @@ from ..decoderonly.configuration_decoderonly import RBLNDecoderOnlyModelConfig
 
 
 class RBLNColQwen2ForRetrievalConfig(RBLNDecoderOnlyModelConfig):
+    """
+    Configuration class for RBLN ColQwen2 models for document retrieval.
+
+    This class extends RBLNModelConfig with specific configurations for ColQwen2 models,
+    including vision tower settings and multi-sequence length support.
+
+    Example usage:
+        ```python
+        from optimum.rbln import RBLNColQwen2ForRetrievalConfig, RBLNColQwen2ForRetrievalConfig
+
+        # Create a configuration object
+        config = RBLNColQwen2ForRetrievalConfig(
+            visual={
+                "max_seq_lens": 6400,
+                "device": 0,
+            },
+            max_seq_len=32_768,
+            tensor_parallel_size=4,
+            device=[0, 1, 2, 3],
+            output_hidden_states=False,
+        )
+
+        # Use the configuration with from_pretrained
+        model = RBLNColQwen2ForRetrieval.from_pretrained(
+            "vidore/colqwen2-v1.0-hf",
+            export=True,
+            rbln_config=config
+        )
+        ```
+    """
+
     submodules = ["visual"]
 
     def __init__(

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/src/optimum/rbln/transformers/models/colqwen2/modeling_colqwen2.py

@@ -51,6 +51,63 @@ from .colqwen2_architecture import ColQwen2LanguageModelWrapper
 
 
 class RBLNColQwen2ForRetrieval(RBLNDecoderOnlyModel):
+    """
+    The ColQwen Model transformer for document retrieval using vision-language models.
+    This model inherits from [`RBLNDecoderOnlyModel`]. Check the superclass documentation for the generic methods the library implements for all its models.
+
+    A class to convert and run pre-trained transformers based `ColQwen2ForRetrieval` model on RBLN devices.
+    It implements the methods to convert a pre-trained transformers `ColQwen2ForRetrieval` model into a RBLN transformer model by:
+
+    - transferring the checkpoint weights of the original into an optimized RBLN graph,
+    - compiling the resulting graph using the RBLN compiler.
+
+    **Configuration:**
+    This model uses [`RBLNColQwen2ForRetrievalConfig`] for configuration. When calling methods like `from_pretrained` or `from_model`,
+    the `rbln_config` parameter should be an instance of [`RBLNColQwen2ForRetrievalConfig`] or a dictionary conforming to its structure.
+
+    See the [`RBLNColQwen2ForRetrievalConfig`] class for all available configuration options.
+
+    Examples:
+        ```python
+        from optimum.rbln import RBLNColQwen2ForRetrieval
+
+        # Using a config dictionary
+        rbln_config = {
+            "visual": {
+                "max_seq_lens": 6400,
+            },
+            "max_seq_len": 32_768,
+            "tensor_parallel_size": 4,
+            "device": [0, 1, 2, 3],
+            "output_hidden_states": False,
+        }
+        model = RBLNColQwen2ForRetrieval.from_pretrained(
+            "vidore/colqwen2-v1.0-hf",
+            export=True,
+            rbln_config=rbln_config
+        )
+
+        # Using a RBLNColQwen2ForRetrievalConfig instance (recommended for type checking)
+        from optimum.rbln import RBLNColQwen2ForRetrievalConfig
+
+        config = RBLNColQwen2ForRetrievalConfig(
+            visual={
+                "max_seq_lens": 6400,
+                "device": 0,
+            },
+            max_seq_len=32_768,
+            tensor_parallel_size=4,
+            device=[0, 1, 2, 3],
+            output_hidden_states=False,
+        )
+        model = RBLNColQwen2ForRetrieval.from_pretrained(
+            "vidore/colqwen2-v1.0-hf",
+            export=True,
+            rbln_config=config
+        )
+        ```
+    """
+
     main_input_name = "inputs_embeds"
     auto_model_class = None
     _rbln_submodules = [

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/tests/test_base.py

@@ -197,16 +197,20 @@ class BaseTest:
                     output = self.model(**inputs)[0]
 
             output = self.postprocess(inputs, output)
-            if self.EXPECTED_OUTPUT:
+            if self.EXPECTED_OUTPUT and self.DEVICE is None:
                 from simphile import jaccard_similarity
 
                 if isinstance(self.EXPECTED_OUTPUT, str):
                     similarity = jaccard_similarity(output, self.EXPECTED_OUTPUT)
-                    self.assertGreater(similarity, 0.9)
+                    self.assertGreater(
+                        similarity, 0.9, msg=f"self.EXPECTED_OUTPUT: {self.EXPECTED_OUTPUT}, output: {output}"
+                    )
                 else:
                     for o, e_o in zip(output, self.EXPECTED_OUTPUT):
                         similarity = jaccard_similarity(o, e_o)
-                        self.assertGreater(similarity, 0.9)
+                        self.assertGreater(
+                            similarity, 0.9, msg=f"self.EXPECTED_OUTPUT: {self.EXPECTED_OUTPUT}, output: {output}"
+                        )
 
         def _inner_test_save_load(self, tmpdir):
             with ContextRblnConfig(create_runtimes=False):

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/tests/test_diffusers.py

@@ -18,7 +18,7 @@ from optimum.rbln import (
     RBLNStableDiffusionXLPipeline,
 )
 
-from .test_base import BaseHubTest, BaseTest
+from .test_base import BaseHubTest, BaseTest, TestLevel
 
 
 class TestSDModel(BaseTest.TestModel, BaseHubTest.TestHub):
@@ -83,6 +83,7 @@ class TestSDXLModel(BaseTest.TestModel):
             },
         }
     }
+    TEST_LEVEL = TestLevel.DISABLED  # Should be enabled after compiler issue is fixed
 
 
 class TestSDImg2ImgModel(BaseTest.TestModel):
@@ -165,6 +166,8 @@ class TestSDXLControlNetModel(BaseTest.TestModel):
         },
     }
 
+    TEST_LEVEL = TestLevel.DISABLED  # Should be enabled after compiler issue is fixed
+
     @classmethod
     def setUpClass(cls):
         controlnet = ControlNetModel.from_pretrained(cls.CONTROLNET_ID)
@@ -262,6 +265,7 @@ class TestKandinskyV22Model(BaseTest.TestModel):
             "decoder_pipe": {"unet": {"batch_size": 2}},
         },
     }
+    TEST_LEVEL = TestLevel.DISABLED  # Should be enabled after compiler issue is fixed
 
     def test_complicate_config(self):
         rbln_config = {
@@ -312,6 +316,7 @@ class TestKandinskyV22Img2ImgModel(BaseTest.TestModel):
             "decoder_pipe": {"unet": {"batch_size": 2}},
         },
     }
+    TEST_LEVEL = TestLevel.DISABLED  # Should be enabled after compiler issue is fixed
 
 
 if __name__ == "__main__":

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/tests/test_llm.py

@@ -99,7 +99,6 @@ class LLMTest:
 class TestMistralForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNMistralForCausalLM
     HF_MODEL_ID = "openaccess-ai-collective/tiny-mistral"
-    EXPECTED_OUTPUT = "watasurescid completionennen Brad completion жеULT ba completion影 Fin сво Regimentixon cabin影 provisions bland"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024, "sliding_window": 512}
 
 
@@ -112,24 +111,21 @@ class TestMistralModel(LLMTest.TestLLMWithoutLMHead):
 class TestQwen2ForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNQwen2ForCausalLM
     HF_MODEL_ID = "Qwen/Qwen2-0.5B-Instruct"
-    EXPECTED_OUTPUT = "?:雨成名ylonclaimer淡elsinki一角一角一角一角一角一角一角一角一角一角一角一角一角"
-    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024}
+    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "layer_types": ["full_attention"], "max_position_embeddings": 1024}
 
 
 class TestQwen2Model(LLMTest.TestLLMWithoutLMHead):
     RBLN_CLASS = RBLNQwen2Model
     HF_MODEL_ID = "Qwen/Qwen2-0.5B-Instruct"
-    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024}
+    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "layer_types": ["full_attention"], "max_position_embeddings": 1024}
 
 
 class TestQwen3ForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNQwen3ForCausalLM
     HF_MODEL_ID = "trl-internal-testing/tiny-Qwen3ForCausalLM"
-    EXPECTED_OUTPUT = (
-        "יל synd Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz Fitz_inventory天河 sanitary中途"
-    )
     HF_CONFIG_KWARGS = {
         "num_hidden_layers": 1,
+        "layer_types": ["full_attention"],
         "max_position_embeddings": 1024,
         "revision": "397c180b0ded9c45c33bbce7f88df86bb2d571d4",
     }
@@ -144,7 +140,7 @@ class TestQwen3Model(LLMTest.TestLLMWithoutLMHead):
     RBLN_AUTO_CLASS = RBLNAutoModel
     RBLN_CLASS = RBLNQwen3Model
     HF_MODEL_ID = "trl-internal-testing/tiny-Qwen3ForCausalLM"
-    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024}
+    HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "layer_types": ["full_attention"], "max_position_embeddings": 1024}
 
 
 class TestQwen3Model_UAM(TestQwen3Model):
@@ -154,7 +150,6 @@ class TestQwen3Model_UAM(TestQwen3Model):
 class TestOPTForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNOPTForCausalLM
     HF_MODEL_ID = "facebook/opt-2.7b"
-    EXPECTED_OUTPUT = "????,,,,,,,,,,,,,,,,"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 2048}
 
 
@@ -168,7 +163,6 @@ class TestLlamaForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNLlamaForCausalLM
     HF_MODEL_ID = "afmck/testing-llama-tiny"
     TEST_LEVEL = TestLevel.ESSENTIAL
-    EXPECTED_OUTPUT = "reress makefable R���� noethetsshss rechoolso�"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024}
 
 
@@ -182,7 +176,6 @@ class TestLlamaForCausalLM_Flash(LLMTest.TestLLM):
     RBLN_CLASS = RBLNLlamaForCausalLM
     HF_MODEL_ID = "afmck/testing-llama-tiny"
     TEST_LEVEL = TestLevel.ESSENTIAL
-    EXPECTED_OUTPUT = "reress makefable R���� noethetsshss rechoolso�"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 8192}
     RBLN_CLASS_KWARGS = {"rbln_config": {"attn_impl": "flash_attn", "kvcache_partition_len": 4096}}
 
@@ -196,19 +189,11 @@ class TestLlamaModel_Flash(LLMTest.TestLLMWithoutLMHead):
 
 class TestLlamaForCausalLM_Multibatch(TestLlamaForCausalLM):
     PROMPT = ["Who are you?", "What is the capital of France?", "What is the capital of Germany?"]
-    EXPECTED_OUTPUT = [
-        "reress makefable R���� noethetss0oss invetetet",
-        "resget makeget makeichget makeichualichual#choolchool accngngngng",
-        "resget makeget makeichget makeichualichual#choolchool accngngngng",
-    ]
     RBLN_CLASS_KWARGS = {"rbln_config": {"batch_size": 3, "decoder_batch_sizes": [3, 2, 1]}}
 
 
 class TestGPT2LMHeadModel(LLMTest.TestLLM):
     RBLN_CLASS = RBLNGPT2LMHeadModel
-    EXPECTED_OUTPUT = (
-        " What kind kind kind kind kind kind kind kind kind kind kind kind kind kind kind kind kind kind kind"
-    )
     HF_MODEL_ID = "openai-community/gpt2"
     HF_CONFIG_KWARGS = {"n_layer": 1, "max_position_embeddings": 1024}
 
@@ -224,7 +209,6 @@ class TestPhiForCausalLM(LLMTest.TestLLM):
 
     # HF_MODEL_ID = "hf-internal-testing/tiny-random-PhiForCausalLM"
     HF_MODEL_ID = "microsoft/phi-2"
-    EXPECTED_OUTPUT = "\nAnswer: Theorettebrates']['<<<urlskolegateezzingrill"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024, "trust_remote_code": True}
 
 
@@ -238,7 +222,6 @@ class TestExaoneForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNExaoneForCausalLM
     # HF_MODEL_ID = "katuni4ka/tiny-random-exaone"
     HF_MODEL_ID = "LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct"
-    EXPECTED_OUTPUT = "????????????????????"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024, "trust_remote_code": True}
 
 
@@ -250,7 +233,6 @@ class TestT5Model(LLMTest.TestLLM):
     # FIXME:: Update to internal once enabled tiny model
     HF_MODEL_ID = "t5-small"
     PROMPT = "summarize: studies have shown that owning a dog is good for you"
-    EXPECTED_OUTPUT = ""
     RBLN_CLASS_KWARGS = {"rbln_config": {"enc_max_seq_len": 512, "dec_max_seq_len": 512}}
     HF_CONFIG_KWARGS = {"num_layers": 1}
 
@@ -284,7 +266,6 @@ class TestBartModel(LLMTest.TestLLM):
     }
     RBLN_CLASS_KWARGS = {"rbln_config": {"enc_max_seq_len": 512, "dec_max_seq_len": 512}}
     PROMPT = "summarize: studies have shown that owning a dog is good for you"
-    EXPECTED_OUTPUT = "InsteadInsteadInsteadHoweverHoweverHoweverAlthoughAlthoughAlthoughWhileWhileWhileAlthoughAlthoughHoweverHoweverManyMany"
     TEST_LEVEL = TestLevel.ESSENTIAL
 
     def get_inputs(self):
@@ -351,7 +332,6 @@ class TestLlavaForConditionalGeneration(LLMTest.TestLLM):
             "language_model": {"use_inputs_embeds": True},
         }
     }
-    EXPECTED_OUTPUT = "ambbrow nur Well chimCore rapideraine Йye questaédédates Ken neu Airport din termeächstthread"
     HF_CONFIG_KWARGS = {"revision": "8ab8bfc820a6bb9e0f8de1ac715f4b53db44e684"}
     HF_CONFIG_KWARGS_PREPROCESSOR = {"revision": "8ab8bfc820a6bb9e0f8de1ac715f4b53db44e684"}
     IS_MULTIMODAL = True
@@ -390,7 +370,6 @@ class TestPegasusModel(LLMTest.TestLLM):
     }
     RBLN_CLASS_KWARGS = {"rbln_config": {"enc_max_seq_len": 512, "dec_max_seq_len": 512}}
     PROMPT = "summarize: studies have shown that owning a dog is good for you"
-    EXPECTED_OUTPUT = "The The The The The The The The The The The The The The The The The The The"
     TEST_LEVEL = TestLevel.ESSENTIAL
 
     def get_inputs(self):
@@ -419,7 +398,6 @@ class TestLlavaNextForConditionalGeneration(LLMTest.TestLLM):
             "language_model": {"use_inputs_embeds": True},
         }
     }
-    EXPECTED_OUTPUT = "entricCallbackavidARYails NotesDAPimil coordFeed Boysaml obligation relay迟 войны sexual Definition Eisen patent"
     HF_CONFIG_KWARGS = {"revision": "21948c1af6a0666e341b6403dc1cbbd5c8900e7d"}
     HF_CONFIG_KWARGS_PREPROCESSOR = {"revision": "21948c1af6a0666e341b6403dc1cbbd5c8900e7d"}
     IS_MULTIMODAL = True
@@ -478,7 +456,6 @@ class TestBlip2ForConditionalGeneration(LLMTest.TestLLM):
     HF_MODEL_ID = "Salesforce/blip2-opt-2.7b"  # No tiny model yet.
     PROMPT = "Question: Describe this image? Answer:"
     RBLN_CLASS_KWARGS = {"rbln_config": {"language_model": {"use_inputs_embeds": True, "max_seq_len": 1024}}}
-    EXPECTED_OUTPUT = "::::::::::::::::::::"
     HF_CONFIG_KWARGS = {}  # Initialize empty to avoid sharing with other classes
     IS_MULTIMODAL = True
 
@@ -559,10 +536,7 @@ class TestQwen2VLForConditionalGeneration(LLMTest.TestLLM):
             "max_seq_len": 32_768,
         }
     }
-    EXPECTED_OUTPUT = " finishing Flight旅馆ジ eventual Rivers Ủy#! Bilderbroker WageאוגוסTransactionإبد.operator офис lives #-}\n voltage"
-    HF_CONFIG_KWARGS = {
-        "num_hidden_layers": 1,
-    }
+    HF_CONFIG_KWARGS = {"num_hidden_layers": 1}
 
     @classmethod
     def setUpClass(cls):
@@ -605,7 +579,6 @@ class TestQwen2_5_VLForConditionalGeneration(LLMTest.TestLLM):
             "max_seq_len": 32_768,
         }
     }
-    EXPECTED_OUTPUT = "讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣讣"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1}
     HF_CONFIG_KWARGS_PREPROCESSOR = {"max_pixels": 64 * 14 * 14}
     IS_MULTIMODAL = True
@@ -636,10 +609,9 @@ class TestQwen2_5_VLForConditionalGeneration(LLMTest.TestLLM):
 class TestGemma3ForConditionalGeneration(LLMTest.TestLLM):
     RBLN_AUTO_CLASS = RBLNAutoModelForImageTextToText
     RBLN_CLASS = RBLNGemma3ForConditionalGeneration
-    HF_MODEL_ID = "trl-internal-testing/tiny-Gemma3ForConditionalGeneration"  # No tiny model yet.
+    HF_MODEL_ID = "trl-internal-testing/tiny-Gemma3ForConditionalGeneration"
     PROMPT = "<bos><start_of_turn>user\n<start_of_image>Describe the image.<end_of_turn>\n<start_of_turn>model\n'"
     RBLN_CLASS_KWARGS = {"rbln_config": {"language_model": {"use_inputs_embeds": True, "kvcache_partition_len": 4096}}}
-    EXPECTED_OUTPUT = " அனுமதி Bryson Earlyheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserheiserिल्म हस्ता"
     HF_CONFIG_KWARGS = {"revision": "e1f4b0516ec80f86ed75c8cb1d45ede72526ad24"}
     HF_CONFIG_KWARGS_PREPROCESSOR = {"revision": "e1f4b0516ec80f86ed75c8cb1d45ede72526ad24"}
     TEST_LEVEL = TestLevel.FULL
@@ -674,14 +646,21 @@ class TestGemma3ForConditionalGeneration(LLMTest.TestLLM):
 class TestGemma3ForCausalLM(LLMTest.TestLLM):
     RBLN_CLASS = RBLNGemma3ForCausalLM
     HF_MODEL_ID = "google/gemma-3-1b-it"
-    EXPECTED_OUTPUT = "1st L L L L L L L L L L L L L L L L L L"
     HF_CONFIG_KWARGS = {
-        "num_hidden_layers": 2,
-        "sliding_window_pattern": 2,
-        "max_position_embeddings": 1024,
         "trust_remote_code": True,
     }
 
+    @classmethod
+    def setUpClass(cls):
+        hf_config = AutoConfig.from_pretrained(cls.HF_MODEL_ID)
+        hf_config.num_hidden_layers = 2
+        hf_config.layer_types = ["full_attention", "sliding_attention"]
+        hf_config.sliding_window_pattern = 2
+        hf_config.max_position_embeddings = 1024
+        kwargs = {"config": hf_config}
+        cls.HF_CONFIG_KWARGS.update(kwargs)
+        return super().setUpClass()
+
 
 class TestLlamaForCausalLM_fp8(LLMTest.TestLLM):
     RBLN_CLASS = RBLNLlamaForCausalLM
@@ -698,7 +677,6 @@ class TestLlamaForCausalLM_fp8(LLMTest.TestLLM):
             "tensor_parallel_size": 1,
         },
     }
-    TEST_LEVEL = TestLevel.DISABLED
 
     def test_generate(self):
         # Cannot generate output with fp8 quantization in ATOM™
@@ -706,8 +684,6 @@ class TestLlamaForCausalLM_fp8(LLMTest.TestLLM):
 
 
 class TestMultiLora(LLMTest.TestLLM):
-    EXPECTED_OUTPUT = " bench_echointon Ebonylica Lennonnings909 norgeZN°Eusanpha701OPSadeleepromtrap乎 Howe"
-
     HF_MODEL_ID = "meta-llama/Llama-3.1-8B-Instruct"
     HF_CONFIG_KWARGS = {"num_hidden_layers": 1, "max_position_embeddings": 1024}
     RBLN_CLASS = RBLNLlamaForCausalLM
@@ -730,6 +706,8 @@ class TestMultiLora(LLMTest.TestLLM):
 
 class TestMultiLora_batch(LLMTest.TestLLM):
     PROMPT = ["Who are you?", "What is the capital of France?"]
+
+    # Should check each output corresponds to each prompt
     EXPECTED_OUTPUT = [
         " bench_echointon Ebonylica Lennonnings909 norgeZN°Eusan倍oloadolen逸 Oaksodian surplusaniem",
         "/topicпідonus343../../../ Mund  Ont ReactionIPAچیIQUE beltーブ204umlu Cortexoisئةτερ",

{optimum_rbln-0.9.2rc0 → optimum_rbln-0.9.2rc1}/tests/test_transformers.py

@@ -404,6 +404,7 @@ class TestColQwen2Model(BaseTest.TestModel):
             "max_seq_len": 32_768,
         }
     }
+    HF_CONFIG_KWARGS = {}  # Initialize empty to avoid sharing with other classes
     HF_CONFIG_KWARGS_PREPROCESSOR = {"max_pixels": 64 * 14 * 14}
 
     @classmethod

optimum_rbln-0.9.2rc0/.github/version.yaml

@@ -1 +0,0 @@
-rebel_compiler_version: 0.9.2.dev164+g0896157e.prod