Enable different torch dtype in sub models (#34873)

* fix

* fix test

* add tests

* add more tests

* fix tests

* supposed to be a torch.dtype test

* handle BC and make fp32 default

src/transformers/configuration_utils.py

@@ -994,8 +994,11 @@ class PretrainedConfig(PushToHubMixin):
         converts torch.dtype to a string of just the type. For example, `torch.float32` get converted into *"float32"*
         string, which can then be stored in the json format.
         """
-        if d.get("torch_dtype", None) is not None and not isinstance(d["torch_dtype"], str):
-            d["torch_dtype"] = str(d["torch_dtype"]).split(".")[1]
+        if d.get("torch_dtype", None) is not None:
+            if isinstance(d["torch_dtype"], dict):
+                d["torch_dtype"] = {k: str(v).split(".")[-1] for k, v in d["torch_dtype"].items()}
+            elif not isinstance(d["torch_dtype"], str):
+                d["torch_dtype"] = str(d["torch_dtype"]).split(".")[1]
         for value in d.values():
             if isinstance(value, dict):
                 self.dict_torch_dtype_to_str(value)

src/transformers/modeling_utils.py

@@ -1312,11 +1312,10 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                 "`PretrainedConfig`. To create a model from a pretrained model use "
                 f"`model = {self.__class__.__name__}.from_pretrained(PRETRAINED_MODEL_NAME)`"
             )
-        # Save config and origin of the pretrained weights if given in model
         if not getattr(config, "_attn_implementation_autoset", False):
-            config = self._autoset_attn_implementation(
-                config, torch_dtype=torch.get_default_dtype(), check_device_map=False
-            )
+            # config usually has a `torch_dtype` but we need the next line for the `no_super_init` tests
+            dtype = config.torch_dtype if hasattr(config, "torch_dtype") else torch.get_default_dtype()
+            config = self._autoset_attn_implementation(config, torch_dtype=dtype, check_device_map=False)
         self.config = config
 
         # for initialization of the loss
@@ -1411,7 +1410,10 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
         # when we init a model from within another model (e.g. VLMs) and dispatch on FA2
         # a warning is raised that dtype should be fp16. Since we never pass dtype from within
         # modeling code, we can try to infer it here same way as done in `from_pretrained`
-        torch_dtype = kwargs.pop("torch_dtype", torch.get_default_dtype())
+        torch_dtype = kwargs.pop("torch_dtype", config.torch_dtype)
+        if isinstance(torch_dtype, str):
+            torch_dtype = getattr(torch, torch_dtype)
+
         use_flash_attention_2 = kwargs.pop("use_flash_attention_2", False)
 
         # override default dtype if needed
@@ -4020,11 +4022,37 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
                             )
                     elif hasattr(torch, torch_dtype):
                         torch_dtype = getattr(torch, torch_dtype)
-                    else:
-                        raise ValueError(
-                            f'`torch_dtype` can be one of: `torch.dtype`, `"auto"` or a string of a valid `torch.dtype`, but received {torch_dtype}'
-                        )
+                        for sub_config_key in config.sub_configs.keys():
+                            sub_config = getattr(config, sub_config_key)
+                            sub_config.torch_dtype = torch_dtype
+                elif isinstance(torch_dtype, torch.dtype):
+                    pass
+                elif isinstance(torch_dtype, dict):
+                    for key, curr_dtype in torch_dtype.items():
+                        if hasattr(config, key):
+                            value = getattr(config, key)
+                            value.torch_dtype = curr_dtype
+                    # main torch dtype for modules that aren't part of any sub-config
+                    torch_dtype = torch_dtype.get("")
+                    config.torch_dtype = torch_dtype
+                    if isinstance(torch_dtype, str) and hasattr(torch, torch_dtype):
+                        torch_dtype = getattr(torch, torch_dtype)
+                    elif torch_dtype is None:
+                        torch_dtype = torch.float32
+                else:
+                    raise ValueError(
+                        f"`torch_dtype` can be one of: `torch.dtype`, `'auto'`, a string of a valid `torch.dtype` or a `dict` with valid `torch_dtype` "
+                        f"for each sub-config in composite configs, but received {torch_dtype}"
+                    )
+
                 dtype_orig = cls._set_default_torch_dtype(torch_dtype)
+            else:
+                # set fp32 as the default dtype for BC
+                default_dtype = str(torch.get_default_dtype()).split(".")[-1]
+                config.torch_dtype = default_dtype
+                for key in config.sub_configs.keys():
+                    value = getattr(config, key)
+                    value.torch_dtype = default_dtype
 
             # Check if `_keep_in_fp32_modules` is not None
             use_keep_in_fp32_modules = (cls._keep_in_fp32_modules is not None) and (

src/transformers/models/chameleon/modeling_chameleon.py

@@ -967,62 +967,6 @@ class ChameleonVQVAEEncoder(nn.Module):
         return last_hidden_state
 
 
-CHAMELEON_VQ_START_DOCSTRING = r"""
-    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
-    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
-    etc.)
-
-    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
-    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
-    and behavior.
-
-    Parameters:
-        config ([`ChameleonVQVAEConfig`]):
-            Model configuration class with all the parameters of the model. Initializing with a config file does not
-            load the weights associated with the model, only the configuration. Check out the
-            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
-"""
-
-
-@add_start_docstrings(
-    """The VQ-VAE model used in Chameleon for encoding/decoding images into discrete tokens.
-    This model follows the "Make-a-scene: Scene-based text-to-image generation with human priors" paper from
-    [ Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman](https://arxiv.org/abs/2203.13131).
-    """,
-    CHAMELEON_VQ_START_DOCSTRING,
-)
-class ChameleonVQVAE(PreTrainedModel):
-    config_class = ChameleonVQVAEConfig
-    _no_split_modules = ["ChameleonVQVAEVectorQuantizer"]
-
-    def _init_weights(self, module):
-        std = self.config.initializer_range
-        if isinstance(module, nn.Embedding):
-            module.weight.data.normal_(mean=0.0, std=std)
-        elif isinstance(module, nn.GroupNorm):
-            module.bias.data.zero_()
-            module.weight.data.fill_(1.0)
-        elif isinstance(module, (nn.Linear, nn.Conv2d)):
-            module.weight.data.normal_(mean=0.0, std=std)
-            if module.bias is not None:
-                module.bias.data.zero_()
-
-    def __init__(self, config: ChameleonVQVAEConfig):
-        super().__init__(config)
-
-        self.encoder = ChameleonVQVAEEncoder(config)
-        self.quantize = ChameleonVQVAEVectorQuantizer(config)
-        self.quant_conv = torch.nn.Conv2d(config.latent_channels, config.embed_dim, 1)
-        self.post_quant_conv = torch.nn.Conv2d(config.embed_dim, config.latent_channels, 1)
-        self.eval()  # Chameleon's VQ model is frozen
-
-    def encode(self, pixel_values: torch.LongTensor):
-        hidden_states = self.encoder(pixel_values)
-        hidden_states = self.quant_conv(hidden_states)
-        quant, emb_loss, indices = self.quantize(hidden_states)
-        return quant, emb_loss, indices
-
-
 class ChameleonImageVocabularyMapping:
     """
     A class for mapping discrete image tokens from VQGAN to BPE tokens.
@@ -1118,6 +1062,62 @@ class ChameleonPreTrainedModel(PreTrainedModel):
                 module.weight.data[module.padding_idx].zero_()
 
 
+CHAMELEON_VQ_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`ChameleonVQVAEConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    """The VQ-VAE model used in Chameleon for encoding/decoding images into discrete tokens.
+    This model follows the "Make-a-scene: Scene-based text-to-image generation with human priors" paper from
+    [ Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman](https://arxiv.org/abs/2203.13131).
+    """,
+    CHAMELEON_VQ_START_DOCSTRING,
+)
+class ChameleonVQVAE(ChameleonPreTrainedModel):
+    config_class = ChameleonVQVAEConfig
+    _no_split_modules = ["ChameleonVQVAEVectorQuantizer"]
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+        elif isinstance(module, nn.GroupNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        elif isinstance(module, (nn.Linear, nn.Conv2d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+
+    def __init__(self, config: ChameleonVQVAEConfig):
+        super().__init__(config)
+
+        self.encoder = ChameleonVQVAEEncoder(config)
+        self.quantize = ChameleonVQVAEVectorQuantizer(config)
+        self.quant_conv = torch.nn.Conv2d(config.latent_channels, config.embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(config.embed_dim, config.latent_channels, 1)
+        self.eval()  # Chameleon's VQ model is frozen
+
+    def encode(self, pixel_values: torch.LongTensor):
+        hidden_states = self.encoder(pixel_values)
+        hidden_states = self.quant_conv(hidden_states)
+        quant, emb_loss, indices = self.quantize(hidden_states)
+        return quant, emb_loss, indices
+
+
 CHAMELEON_INPUTS_DOCSTRING = r"""
     Args:
         input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
@@ -1211,7 +1211,7 @@ class ChameleonModel(ChameleonPreTrainedModel):
             [decoder_layer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
         )
         self.norm = ChameleonRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
-        self.vqmodel = ChameleonVQVAE(config.vq_config)
+        self.vqmodel = ChameleonVQVAE._from_config(config.vq_config)
         self.gradient_checkpointing = False
 
         # Initialize weights and apply final processing

tests/models/qwen2_vl/test_modeling_qwen2_vl.py

@@ -227,6 +227,7 @@ class Qwen2VLModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCas
     pipeline_model_mapping = {"image-text-to-text": Qwen2VLForConditionalGeneration}
     test_pruning = False
     test_head_masking = False
+    _is_composite = True
 
     def setUp(self):
         self.model_tester = Qwen2VLVisionText2TextModelTester(self)

tests/utils/test_modeling_utils.py

@@ -37,6 +37,7 @@ from transformers import (
     AutoModel,
     AutoModelForImageClassification,
     AutoModelForSequenceClassification,
+    LlavaForConditionalGeneration,
     OwlViTForObjectDetection,
     PretrainedConfig,
     is_torch_available,
@@ -300,6 +301,7 @@ TINY_T5 = "patrickvonplaten/t5-tiny-random"
 TINY_BERT_FOR_TOKEN_CLASSIFICATION = "hf-internal-testing/tiny-bert-for-token-classification"
 TINY_MISTRAL = "hf-internal-testing/tiny-random-MistralForCausalLM"
 TINY_IMAGE_CLASSIF = "hf-internal-testing/tiny-random-SiglipForImageClassification"
+TINY_LLAVA = "hf-internal-testing/tiny-random-LlavaForConditionalGeneration"
 
 LOG = logging.get_logger(__name__)
 
@@ -460,6 +462,59 @@ class ModelUtilsTest(TestCasePlus):
         with self.assertRaises(ValueError):
             model = AutoModel.from_pretrained(TINY_T5, torch_dtype="int64")
 
+    def test_model_from_config_torch_dtype_composite(self):
+        """
+        Test that from_pretrained works with torch_dtype being as a dict per each sub-config in composite config
+        """
+        # should be able to set torch_dtype as a simple string and the model loads it correctly
+        model = LlavaForConditionalGeneration.from_pretrained(TINY_LLAVA, torch_dtype="float32")
+        self.assertEqual(model.language_model.dtype, torch.float32)
+        self.assertEqual(model.vision_tower.dtype, torch.float32)
+
+        model = LlavaForConditionalGeneration.from_pretrained(TINY_LLAVA, torch_dtype="float16")
+        self.assertEqual(model.language_model.dtype, torch.float16)
+        self.assertEqual(model.vision_tower.dtype, torch.float16)
+
+        # should be able to set torch_dtype as a dict for each sub-config
+        model = LlavaForConditionalGeneration.from_pretrained(
+            TINY_LLAVA, torch_dtype={"text_config": "float32", "vision_config": "float16", "": "bfloat16"}
+        )
+        self.assertEqual(model.language_model.dtype, torch.float32)
+        self.assertEqual(model.vision_tower.dtype, torch.float16)
+        self.assertEqual(model.multi_modal_projector.linear_1.weight.dtype, torch.bfloat16)
+
+        # should be able to set the values as torch.dtype (not str)
+        model = LlavaForConditionalGeneration.from_pretrained(
+            TINY_LLAVA, torch_dtype={"text_config": torch.float32, "vision_config": torch.float16, "": torch.bfloat16}
+        )
+        self.assertEqual(model.language_model.dtype, torch.float32)
+        self.assertEqual(model.vision_tower.dtype, torch.float16)
+        self.assertEqual(model.multi_modal_projector.linear_1.weight.dtype, torch.bfloat16)
+
+        # should be able to set the values in configs directly and pass it to `from_pretrained`
+        config = copy.deepcopy(model.config)
+        config.text_config.torch_dtype = torch.float32
+        config.vision_config.torch_dtype = torch.bfloat16
+        config.torch_dtype = torch.float16
+        model = LlavaForConditionalGeneration.from_pretrained(TINY_LLAVA, config=config, torch_dtype="auto")
+        self.assertEqual(model.language_model.dtype, torch.float32)
+        self.assertEqual(model.vision_tower.dtype, torch.bfloat16)
+        self.assertEqual(model.multi_modal_projector.linear_1.weight.dtype, torch.float16)
+
+        # but if the model has `_keep_in_fp32_modules` then those modules should be in fp32 no matter what
+        LlavaForConditionalGeneration._keep_in_fp32_modules = ["multi_modal_projector"]
+        model = LlavaForConditionalGeneration.from_pretrained(TINY_LLAVA, config=config, torch_dtype="auto")
+        self.assertEqual(model.language_model.dtype, torch.float32)
+        self.assertEqual(model.vision_tower.dtype, torch.bfloat16)
+        self.assertEqual(model.multi_modal_projector.linear_1.weight.dtype, torch.float32)
+
+        # torch.set_default_dtype() supports only float dtypes, so will fail with non-float type
+        with self.assertRaises(ValueError):
+            model = LlavaForConditionalGeneration.from_pretrained(TINY_LLAVA, torch_dtype="int64")
+            model = LlavaForConditionalGeneration.from_pretrained(
+                TINY_LLAVA, torch_dtype={"text_config": "float32", "vision_config": "int64", "": "float16"}
+            )
+
     @require_torch
     def test_model_from_pretrained_meta_device(self):
         def is_on_meta(model_id, dtype):