[WIP] Pixtral Text Config

docs/source/en/model_doc/pixtral.md

@@ -78,6 +78,10 @@ output = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up
 
 [[autodoc]] PixtralVisionConfig
 
+## PixtralTextConfig
+
+[[autodoc]] PixtralTextConfig
+
 ## PixtralVisionModel
 
 [[autodoc]] PixtralVisionModel

src/transformers/__init__.py

@@ -700,7 +700,7 @@
         "Pix2StructTextConfig",
         "Pix2StructVisionConfig",
     ],
-    "models.pixtral": ["PixtralProcessor", "PixtralVisionConfig"],
+    "models.pixtral": ["PixtralProcessor", "PixtralVisionConfig", "PixtralTextConfig"],
     "models.plbart": ["PLBartConfig"],
     "models.poolformer": ["PoolFormerConfig"],
     "models.pop2piano": ["Pop2PianoConfig"],

src/transformers/models/auto/configuration_auto.py

@@ -232,6 +232,7 @@
         ("phimoe", "PhimoeConfig"),
         ("pix2struct", "Pix2StructConfig"),
         ("pixtral", "PixtralVisionConfig"),
+        ("pixtral_text", "PixtralTextConfig"),
         ("plbart", "PLBartConfig"),
         ("poolformer", "PoolFormerConfig"),
         ("pop2piano", "Pop2PianoConfig"),
@@ -574,6 +575,7 @@
         ("phobert", "PhoBERT"),
         ("pix2struct", "Pix2Struct"),
         ("pixtral", "Pixtral"),
+        ("pixtral_text", "PixtralMistral"),
         ("plbart", "PLBart"),
         ("poolformer", "PoolFormer"),
         ("pop2piano", "Pop2Piano"),
@@ -740,6 +742,7 @@
         ("chinese_clip_vision_model", "chinese_clip"),
         ("rt_detr_resnet", "rt_detr"),
         ("granitevision", "llava_next"),
+        ("pixtral_text", "pixtral"),
     ]
 )
 

src/transformers/models/auto/modeling_auto.py

@@ -555,6 +555,7 @@
         ("phi", "PhiForCausalLM"),
         ("phi3", "Phi3ForCausalLM"),
         ("phimoe", "PhimoeForCausalLM"),
+        ("pixtral_text", "MistralForCausalLM"),
         ("plbart", "PLBartForCausalLM"),
         ("prophetnet", "ProphetNetForCausalLM"),
         ("qdqbert", "QDQBertLMHeadModel"),

src/transformers/models/pixtral/configuration_pixtral.py

@@ -14,6 +14,7 @@
 """Pixtral model configuration"""
 
 from ...configuration_utils import PretrainedConfig
+from ...models.mistral.configuration_mistral import MistralConfig
 from ...utils import logging
 
 
@@ -103,4 +104,116 @@ def __init__(
         self.initializer_range = initializer_range
 
 
-__all__ = ["PixtralVisionConfig"]
+class PixtralTextConfig(MistralConfig):
+    r"""
+    TODO
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 32000):
+            Vocabulary size of the Mistral model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`MistralModel`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 14336):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `8`.
+        head_dim (`int`, *optional*, defaults to `hidden_size // num_attention_heads`):
+            The attention head dimension.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to `4096*32`):
+            The maximum sequence length that this model might ever be used with. Mistral's sliding window attention
+            allows sequence of up to 4096*32 tokens.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            The id of the padding token.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the "end-of-sequence" token.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        sliding_window (`int`, *optional*, defaults to 4096):
+            Sliding window attention window size. If not specified, will default to `4096`.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+
+    ```python
+    >>> TODO
+    ```"""
+
+    model_type = "pixtral_text"
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=14336,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=8,
+        head_dim=None,
+        hidden_act="silu",
+        max_position_embeddings=4096 * 32,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        sliding_window=4096,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+        self.head_dim = head_dim  # as opposed to MistralConfig, do not auto-populate
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+
+        PretrainedConfig.__init__(
+            self,
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["PixtralVisionConfig", "PixtralTextConfig"]

tests/models/llava/test_configuration_llava.py

@@ -0,0 +1,71 @@
+import tempfile
+import unittest
+
+from transformers import LlavaConfig
+
+
+class LlavaConfigTest(unittest.TestCase):
+    def test_llava_reload(self):
+        """
+        Simple test for reloading default llava configs
+        """
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            config = LlavaConfig()
+            config.save_pretrained(tmp_dir)
+
+            reloaded = LlavaConfig.from_pretrained(tmp_dir)
+            assert config.to_dict() == reloaded.to_dict()
+
+    def test_pixtral_reload(self):
+        """
+        Simple test for reloading pixtral configs
+        """
+        vision_config = {
+            "model_type": "pixtral",
+            "head_dim": 64,
+            "hidden_act": "silu",
+            "image_size": 1024,
+            "is_composition": True,
+            "patch_size": 16,
+            "rope_theta": 10000.0,
+            "tie_word_embeddings": False,
+        }
+
+        text_config = {
+            # "model_type": "mistral",
+            "model_type": "pixtral_text",
+            "hidden_size": 5120,
+            "head_dim": 128,
+            "num_attention_heads": 32,
+            "intermediate_size": 14336,
+            "is_composition": True,
+            "max_position_embeddings": 1024000,
+            "num_hidden_layers": 40,
+            "num_key_value_heads": 8,
+            "rms_norm_eps": 1e-05,
+            "rope_theta": 1000000000.0,
+            "sliding_window": None,
+            "vocab_size": 131072,
+        }
+
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            config = LlavaConfig(vision_config=vision_config, text_config=text_config)
+            config.save_pretrained(tmp_dir)
+
+            reloaded = LlavaConfig.from_pretrained(tmp_dir)
+            assert config.to_dict() == reloaded.to_dict()
+
+    def test_arbitrary_reload(self):
+        """
+        Simple test for reloading arbirarily composed subconfigs
+        """
+        default_values = LlavaConfig().to_dict()
+        default_values["vision_config"]["model_type"] = "qwen2_vl"
+        default_values["text_config"]["model_type"] = "opt"
+
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            config = LlavaConfig(**default_values)
+            config.save_pretrained(tmp_dir)
+
+            reloaded = LlavaConfig.from_pretrained(tmp_dir)
+            assert config.to_dict() == reloaded.to_dict()

utils/check_table.py

@@ -180,6 +180,7 @@ def _center_text(text: str, width: int) -> str:
     "CLIPVisionModel",
     "Qwen2AudioEncoder",
     "SiglipVisionModel",
+    "PixtralMistral",  # not a real model
 ]