Skip to content

CLIP (Contrastive Language–Image Pre-training)

CLIP (Contrastive Language–Image Pre-training) is a neural network architecture that learns visual concepts from natural language supervision. It is trained on a large dataset of image-text pairs to create a unified vision-language model that can understand both images and text in a shared semantic space.

CLIP consists of two main components: 1. A vision encoder (Vision Transformer) that processes images into visual features 2. A text encoder (Transformer) that processes text into textual features

The model is trained using contrastive learning, where it learns to maximize the cosine similarity between the embeddings of matching image-text pairs while minimizing it for non-matching pairs. This allows CLIP to perform zero-shot classification by comparing image embeddings with text embeddings of potential labels.

CLIP was introduced in the paper "Learning Transferable Visual Models From Natural Language Supervision" and has shown remarkable zero-shot generalization capabilities across a wide range of visual classification tasks. The CLIP model combines a Vision Transformer and a Text Transformer to learn joint representations of images and text. It is trained to maximize the similarity between matching image-text pairs while minimizing similarity between non-matching pairs.

jimm.models.clip.CLIPVisionModel

Bases: Module

Source code in src/jimm/models/clip.py 
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
class CLIPVisionModel(nnx.Module):
    def __init__(
        self,
        image_resolution: int,
        vision_layers: int,
        vision_width: int,
        vision_patch_size: int,
        transformer_width: int,
        use_gradient_checkpointing: bool = False,
        rngs: rnglib.Rngs = nnx.Rngs(0),
        dtype: DTypeLike = jnp.float32,
        param_dtype: DTypeLike = jnp.float32,
        mesh: Mesh | None = None,
    ):
        """
        Initialize the Vision Encoder with projection.

        Args:
            image_resolution (int): The resolution of the input images.
            vision_layers (int): The number of layers in the vision transformer.
            vision_width (int): The width of the vision transformer.
            vision_patch_size (int): The patch size of the vision transformer.
            transformer_width (int): The output dimension after projection.
            use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
            rngs (rnglib.Rngs): The random number generator state. Defaults to nnx.Rngs(0).
            dtype (DTypeLike): The data type for computations. Defaults to jnp.float32.
            param_dtype (DTypeLike): The data type for parameters. Defaults to jnp.float32.
            mesh (Mesh | None): The device mesh for parameter sharding. Defaults to None.
        """
        self.vision_layers = vision_layers
        self.vision_width = vision_width
        self.vision_patch_size = vision_patch_size
        self.transformer_width = transformer_width
        self.dtype = dtype

        vision_heads = vision_width // 64

        self.encoder = VisionTransformerBase(
            img_size=image_resolution,
            patch_size=vision_patch_size,
            in_channels=3,
            hidden_size=vision_width,
            num_layers=vision_layers,
            num_heads=vision_heads,
            mlp_dim=vision_width * 4,
            use_pre_norm=True,
            use_patch_bias=False,
            use_quick_gelu=True,
            use_gradient_checkpointing=use_gradient_checkpointing,
            pooling_type="CLS",
            layernorm_epsilon=1e-5,
            dtype=dtype,
            param_dtype=param_dtype,
            mesh=mesh,
            rngs=rngs,
        )
        self.visual_projection = nnx.Linear(
            vision_width,
            transformer_width,
            use_bias=False,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
            kernel_init=sharded_init(nnx.initializers.xavier_uniform(), P(None, "model"), mesh),
        )

    def __call__(self, image: Float[Array, "batch height width channels"], do_projection: bool = True) -> Float[Array, "batch vision_width_or_transformer_width"]:
        """
        Encode images into embeddings.

        Args:
            image (Float[Array, "batch height width channels"]): Batch of input images.
            do_projection (bool): Whether to apply the visual projection layer. Defaults to True.

        Returns:
            Float[Array, "batch vision_width_or_transformer_width"]: Image embeddings.
            Shape depends on do_projection: vision_width if False, transformer_width if True.
        """
        features = self.encoder(image)
        if do_projection:
            return self.visual_projection(features)
        return features

    @classmethod
    def from_pretrained(
        cls,
        model_name_or_path: str,
        use_pytorch: bool = False,
        mesh: Mesh | None = None,
        dtype: DTypeLike = jnp.float32,
        param_dtype: DTypeLike = jnp.float32,
        use_gradient_checkpointing: bool = False,
        rngs: rnglib.Rngs = nnx.Rngs(0),
    ) -> "CLIPVisionModel":
        """Load a pretrained vision encoder from a CLIP checkpoint.

        Args:
            model_name_or_path (str): Path to local weights or HuggingFace model ID.
            use_pytorch (bool): Whether to load from PyTorch weights. Defaults to False.
            mesh (Mesh | None): Optional device mesh for parameter sharding. Defaults to None.
            dtype (DTypeLike): Data type for computations. Defaults to jnp.float32.
            param_dtype (DTypeLike): Data type for parameters. Defaults to jnp.float32.
            use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
            rngs (rnglib.Rngs): Random number generator keys. Defaults to nnx.Rngs(0).

        Returns:
            CLIPVisionModel: Pretrained CLIP vision model
        """
        params_fstate, config_dict = load_params_and_config(model_name_or_path, use_pytorch)

        config: dict[str, Any] = config_dict

        if config == {}:
            if not use_pytorch:
                text_hidden_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[1]

                vision_hidden_size = params_fstate["vision_model.embeddings.class_embedding"].shape[0]
                vision_patch_size = params_fstate["vision_model.embeddings.patch_embedding.weight"].shape[2]
                vision_image_size = int((params_fstate["vision_model.embeddings.position_embedding.weight"].shape[0] - 1) ** 0.5) * vision_patch_size

                vision_num_layers = 0
                for k_param in params_fstate:
                    if k_param.startswith("vision_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                        layer_idx = int(k_param.split(".")[3])
                        vision_num_layers = max(vision_num_layers, layer_idx + 1)

                config = {
                    "text_config": {
                        "hidden_size": text_hidden_size,
                    },
                    "vision_config": {
                        "hidden_size": vision_hidden_size,
                        "num_attention_heads": vision_hidden_size // 64,
                        "num_hidden_layers": vision_num_layers,
                        "image_size": vision_image_size,
                        "patch_size": vision_patch_size,
                    },
                }
            else:
                raise ValueError(f"Configuration could not be loaded for PyTorch model {model_name_or_path}")

        text_config = config["text_config"]
        vision_config = config["vision_config"]

        vision_model = cls(
            image_resolution=vision_config["image_size"],
            vision_layers=vision_config["num_hidden_layers"],
            vision_width=vision_config["hidden_size"],
            vision_patch_size=vision_config["patch_size"],
            transformer_width=text_config["hidden_size"],
            use_gradient_checkpointing=use_gradient_checkpointing,
            mesh=mesh,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
        )

        flax_model_params_fstate = dict(nnx.to_flat_state(nnx.state(vision_model, nnx.Param)))

        vision_mapping_list = [
            (("encoder", "cls_token"), ("vision_model", "embeddings", "class_embedding")),
            (("encoder", "position_embeddings"), ("vision_model", "embeddings", "position_embedding", "weight")),
            (("encoder", "patch_embeddings", "kernel"), ("vision_model", "embeddings", "patch_embedding", "weight")),
            (("encoder", "ln_pre", "scale"), ("vision_model", "pre_layrnorm", "weight")),
            (("encoder", "ln_pre", "bias"), ("vision_model", "pre_layrnorm", "bias")),
            (("encoder", "ln_post", "scale"), ("vision_model", "post_layernorm", "weight")),
            (("encoder", "ln_post", "bias"), ("vision_model", "post_layernorm", "bias")),
            (("visual_projection", "kernel"), ("visual_projection", "weight")),
        ]

        for i in range(vision_config["num_hidden_layers"]):
            flax_base = ("encoder", "encoder", "layers", i)
            hf_base = ("vision_model", "encoder", "layers", str(i))

            vision_mapping_list.extend(
                [
                    (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                    (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                    (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                    (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                    (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                    (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                    (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                    (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                    (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                    (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                    (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                    (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                    (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                    (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                    (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                    (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
                ]
            )

        vision_projection_mapping = [
            (("visual_projection", "kernel"), ("visual_projection", "weight")),
        ]
        params_name_mapping = dict(vision_mapping_list + vision_projection_mapping)
        nonvisited = set(flax_model_params_fstate.keys())

        for flax_dst_key_tuple, hf_src_key_tuple in params_name_mapping.items():
            hf_src_key_as_string = ".".join(hf_src_key_tuple)

            nonvisited.discard(flax_dst_key_tuple)
            src_value = params_fstate[hf_src_key_as_string]
            dst_value_obj = flax_model_params_fstate[flax_dst_key_tuple]

            if flax_dst_key_tuple == ("encoder", "patch_embeddings", "kernel"):
                src_value = jnp.transpose(src_value, (2, 3, 1, 0))
            elif flax_dst_key_tuple == ("encoder", "cls_token"):
                src_value = src_value.reshape(1, 1, -1)
            elif flax_dst_key_tuple == ("encoder", "position_embeddings"):
                src_value = src_value.reshape(1, src_value.shape[0], src_value.shape[1])
            elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((hidden_size, num_heads, head_dim))
            elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim))
            elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim, hidden_size))
            elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
                src_value = jnp.transpose(src_value, (1, 0))

            if src_value.shape != dst_value_obj.value.shape:
                raise ValueError(f"Shape mismatch for {flax_dst_key_tuple} vs {hf_src_key_as_string}: {dst_value_obj.value.shape} (expected) != {src_value.shape} (actual)")

            src_value = src_value.astype(param_dtype)
            dst_value_obj.value = src_value

        nnx.update(vision_model, nnx.from_flat_state(flax_model_params_fstate))

        known_buffer_keys = {("encoder", "vision_position_ids"), ("visual_projection",)}
        unexpected_nonvisited = nonvisited - known_buffer_keys
        if unexpected_nonvisited:
            print(f"Warning: Some CLIPVisionModel parameters were not loaded: {sorted(list(unexpected_nonvisited))}")

        if ("visual_projection", "kernel") not in flax_model_params_fstate and "visual_projection.weight" in params_fstate:
            vision_model.visual_projection.kernel.value = jnp.transpose(params_fstate["visual_projection.weight"], (1, 0)).astype(param_dtype)

        return vision_model

__call__(image, do_projection=True)

Encode images into embeddings.

Parameters:

Name Type Description Default
image Float[Array, 'batch height width channels']

Batch of input images.

 required
do_projection bool

Whether to apply the visual projection layer. Defaults to True.

 True

Returns:

Type Description
Float[Array, 'batch vision_width_or_transformer_width']

Float[Array, "batch vision_width_or_transformer_width"]: Image embeddings.
Float[Array, 'batch vision_width_or_transformer_width']

Shape depends on do_projection: vision_width if False, transformer_width if True.
Source code in src/jimm/models/clip.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
def __call__(self, image: Float[Array, "batch height width channels"], do_projection: bool = True) -> Float[Array, "batch vision_width_or_transformer_width"]:
    """
    Encode images into embeddings.

    Args:
        image (Float[Array, "batch height width channels"]): Batch of input images.
        do_projection (bool): Whether to apply the visual projection layer. Defaults to True.

    Returns:
        Float[Array, "batch vision_width_or_transformer_width"]: Image embeddings.
        Shape depends on do_projection: vision_width if False, transformer_width if True.
    """
    features = self.encoder(image)
    if do_projection:
        return self.visual_projection(features)
    return features

__init__(image_resolution, vision_layers, vision_width, vision_patch_size, transformer_width, use_gradient_checkpointing=False, rngs=nnx.Rngs(0), dtype=jnp.float32, param_dtype=jnp.float32, mesh=None)

Initialize the Vision Encoder with projection.

Parameters:

Name Type Description Default
image_resolution int

The resolution of the input images.

 required
vision_layers int

The number of layers in the vision transformer.

 required
vision_width int

The width of the vision transformer.

 required
vision_patch_size int

The patch size of the vision transformer.

 required
transformer_width int

The output dimension after projection.

 required
use_gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
rngs Rngs

The random number generator state. Defaults to nnx.Rngs(0).

 Rngs(0)
dtype DTypeLike

The data type for computations. Defaults to jnp.float32.

 float32
param_dtype DTypeLike

The data type for parameters. Defaults to jnp.float32.

 float32
mesh Mesh | None

The device mesh for parameter sharding. Defaults to None.

 None
Source code in src/jimm/models/clip.py 
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def __init__(
    self,
    image_resolution: int,
    vision_layers: int,
    vision_width: int,
    vision_patch_size: int,
    transformer_width: int,
    use_gradient_checkpointing: bool = False,
    rngs: rnglib.Rngs = nnx.Rngs(0),
    dtype: DTypeLike = jnp.float32,
    param_dtype: DTypeLike = jnp.float32,
    mesh: Mesh | None = None,
):
    """
    Initialize the Vision Encoder with projection.

    Args:
        image_resolution (int): The resolution of the input images.
        vision_layers (int): The number of layers in the vision transformer.
        vision_width (int): The width of the vision transformer.
        vision_patch_size (int): The patch size of the vision transformer.
        transformer_width (int): The output dimension after projection.
        use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
        rngs (rnglib.Rngs): The random number generator state. Defaults to nnx.Rngs(0).
        dtype (DTypeLike): The data type for computations. Defaults to jnp.float32.
        param_dtype (DTypeLike): The data type for parameters. Defaults to jnp.float32.
        mesh (Mesh | None): The device mesh for parameter sharding. Defaults to None.
    """
    self.vision_layers = vision_layers
    self.vision_width = vision_width
    self.vision_patch_size = vision_patch_size
    self.transformer_width = transformer_width
    self.dtype = dtype

    vision_heads = vision_width // 64

    self.encoder = VisionTransformerBase(
        img_size=image_resolution,
        patch_size=vision_patch_size,
        in_channels=3,
        hidden_size=vision_width,
        num_layers=vision_layers,
        num_heads=vision_heads,
        mlp_dim=vision_width * 4,
        use_pre_norm=True,
        use_patch_bias=False,
        use_quick_gelu=True,
        use_gradient_checkpointing=use_gradient_checkpointing,
        pooling_type="CLS",
        layernorm_epsilon=1e-5,
        dtype=dtype,
        param_dtype=param_dtype,
        mesh=mesh,
        rngs=rngs,
    )
    self.visual_projection = nnx.Linear(
        vision_width,
        transformer_width,
        use_bias=False,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
        kernel_init=sharded_init(nnx.initializers.xavier_uniform(), P(None, "model"), mesh),
    )

from_pretrained(model_name_or_path, use_pytorch=False, mesh=None, dtype=jnp.float32, param_dtype=jnp.float32, use_gradient_checkpointing=False, rngs=nnx.Rngs(0)) classmethod

Load a pretrained vision encoder from a CLIP checkpoint.

Parameters:

Name Type Description Default
model_name_or_path str

Path to local weights or HuggingFace model ID.

 required
use_pytorch bool

Whether to load from PyTorch weights. Defaults to False.

 False
mesh Mesh | None

Optional device mesh for parameter sharding. Defaults to None.

 None
dtype DTypeLike

Data type for computations. Defaults to jnp.float32.

 float32
param_dtype DTypeLike

Data type for parameters. Defaults to jnp.float32.

 float32
use_gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
rngs Rngs

Random number generator keys. Defaults to nnx.Rngs(0).

 Rngs(0)

Returns:

Name Type Description
CLIPVisionModel CLIPVisionModel

Pretrained CLIP vision model
Source code in src/jimm/models/clip.py 
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
@classmethod
def from_pretrained(
    cls,
    model_name_or_path: str,
    use_pytorch: bool = False,
    mesh: Mesh | None = None,
    dtype: DTypeLike = jnp.float32,
    param_dtype: DTypeLike = jnp.float32,
    use_gradient_checkpointing: bool = False,
    rngs: rnglib.Rngs = nnx.Rngs(0),
) -> "CLIPVisionModel":
    """Load a pretrained vision encoder from a CLIP checkpoint.

    Args:
        model_name_or_path (str): Path to local weights or HuggingFace model ID.
        use_pytorch (bool): Whether to load from PyTorch weights. Defaults to False.
        mesh (Mesh | None): Optional device mesh for parameter sharding. Defaults to None.
        dtype (DTypeLike): Data type for computations. Defaults to jnp.float32.
        param_dtype (DTypeLike): Data type for parameters. Defaults to jnp.float32.
        use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
        rngs (rnglib.Rngs): Random number generator keys. Defaults to nnx.Rngs(0).

    Returns:
        CLIPVisionModel: Pretrained CLIP vision model
    """
    params_fstate, config_dict = load_params_and_config(model_name_or_path, use_pytorch)

    config: dict[str, Any] = config_dict

    if config == {}:
        if not use_pytorch:
            text_hidden_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[1]

            vision_hidden_size = params_fstate["vision_model.embeddings.class_embedding"].shape[0]
            vision_patch_size = params_fstate["vision_model.embeddings.patch_embedding.weight"].shape[2]
            vision_image_size = int((params_fstate["vision_model.embeddings.position_embedding.weight"].shape[0] - 1) ** 0.5) * vision_patch_size

            vision_num_layers = 0
            for k_param in params_fstate:
                if k_param.startswith("vision_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                    layer_idx = int(k_param.split(".")[3])
                    vision_num_layers = max(vision_num_layers, layer_idx + 1)

            config = {
                "text_config": {
                    "hidden_size": text_hidden_size,
                },
                "vision_config": {
                    "hidden_size": vision_hidden_size,
                    "num_attention_heads": vision_hidden_size // 64,
                    "num_hidden_layers": vision_num_layers,
                    "image_size": vision_image_size,
                    "patch_size": vision_patch_size,
                },
            }
        else:
            raise ValueError(f"Configuration could not be loaded for PyTorch model {model_name_or_path}")

    text_config = config["text_config"]
    vision_config = config["vision_config"]

    vision_model = cls(
        image_resolution=vision_config["image_size"],
        vision_layers=vision_config["num_hidden_layers"],
        vision_width=vision_config["hidden_size"],
        vision_patch_size=vision_config["patch_size"],
        transformer_width=text_config["hidden_size"],
        use_gradient_checkpointing=use_gradient_checkpointing,
        mesh=mesh,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
    )

    flax_model_params_fstate = dict(nnx.to_flat_state(nnx.state(vision_model, nnx.Param)))

    vision_mapping_list = [
        (("encoder", "cls_token"), ("vision_model", "embeddings", "class_embedding")),
        (("encoder", "position_embeddings"), ("vision_model", "embeddings", "position_embedding", "weight")),
        (("encoder", "patch_embeddings", "kernel"), ("vision_model", "embeddings", "patch_embedding", "weight")),
        (("encoder", "ln_pre", "scale"), ("vision_model", "pre_layrnorm", "weight")),
        (("encoder", "ln_pre", "bias"), ("vision_model", "pre_layrnorm", "bias")),
        (("encoder", "ln_post", "scale"), ("vision_model", "post_layernorm", "weight")),
        (("encoder", "ln_post", "bias"), ("vision_model", "post_layernorm", "bias")),
        (("visual_projection", "kernel"), ("visual_projection", "weight")),
    ]

    for i in range(vision_config["num_hidden_layers"]):
        flax_base = ("encoder", "encoder", "layers", i)
        hf_base = ("vision_model", "encoder", "layers", str(i))

        vision_mapping_list.extend(
            [
                (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
            ]
        )

    vision_projection_mapping = [
        (("visual_projection", "kernel"), ("visual_projection", "weight")),
    ]
    params_name_mapping = dict(vision_mapping_list + vision_projection_mapping)
    nonvisited = set(flax_model_params_fstate.keys())

    for flax_dst_key_tuple, hf_src_key_tuple in params_name_mapping.items():
        hf_src_key_as_string = ".".join(hf_src_key_tuple)

        nonvisited.discard(flax_dst_key_tuple)
        src_value = params_fstate[hf_src_key_as_string]
        dst_value_obj = flax_model_params_fstate[flax_dst_key_tuple]

        if flax_dst_key_tuple == ("encoder", "patch_embeddings", "kernel"):
            src_value = jnp.transpose(src_value, (2, 3, 1, 0))
        elif flax_dst_key_tuple == ("encoder", "cls_token"):
            src_value = src_value.reshape(1, 1, -1)
        elif flax_dst_key_tuple == ("encoder", "position_embeddings"):
            src_value = src_value.reshape(1, src_value.shape[0], src_value.shape[1])
        elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
            src_value = jnp.transpose(src_value, (1, 0))
            num_heads = vision_config["hidden_size"] // 64
            hidden_size = vision_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((hidden_size, num_heads, head_dim))
        elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
            num_heads = vision_config["hidden_size"] // 64
            hidden_size = vision_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((num_heads, head_dim))
        elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
            src_value = jnp.transpose(src_value, (1, 0))
            num_heads = vision_config["hidden_size"] // 64
            hidden_size = vision_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((num_heads, head_dim, hidden_size))
        elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
            src_value = jnp.transpose(src_value, (1, 0))

        if src_value.shape != dst_value_obj.value.shape:
            raise ValueError(f"Shape mismatch for {flax_dst_key_tuple} vs {hf_src_key_as_string}: {dst_value_obj.value.shape} (expected) != {src_value.shape} (actual)")

        src_value = src_value.astype(param_dtype)
        dst_value_obj.value = src_value

    nnx.update(vision_model, nnx.from_flat_state(flax_model_params_fstate))

    known_buffer_keys = {("encoder", "vision_position_ids"), ("visual_projection",)}
    unexpected_nonvisited = nonvisited - known_buffer_keys
    if unexpected_nonvisited:
        print(f"Warning: Some CLIPVisionModel parameters were not loaded: {sorted(list(unexpected_nonvisited))}")

    if ("visual_projection", "kernel") not in flax_model_params_fstate and "visual_projection.weight" in params_fstate:
        vision_model.visual_projection.kernel.value = jnp.transpose(params_fstate["visual_projection.weight"], (1, 0)).astype(param_dtype)

    return vision_model

jimm.models.clip.CLIP

Bases: Module

Source code in src/jimm/models/clip.py 
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
class CLIP(nnx.Module):
    def __init__(
        self,
        image_resolution: int,
        vision_layers: int,
        vision_width: int,
        vision_patch_size: int,
        context_length: int,
        vocab_size: int,
        transformer_width: int,
        transformer_heads: int,
        transformer_layers: int,
        use_gradient_checkpointing: bool = False,
        rngs: rnglib.Rngs = nnx.Rngs(0),
        dtype: DTypeLike = jnp.float32,
        param_dtype: DTypeLike = jnp.float32,
        mesh: Mesh | None = None,
    ):
        """
        Initialize the CLIP model.

        Args:
            image_resolution (int): The resolution of the input images.
            vision_layers (int): The number of layers in the vision transformer.
            vision_width (int): The width of the vision transformer.
            vision_patch_size (int): The patch size of the vision transformer.
            context_length (int): The length of the context.
            vocab_size (int): The size of the vocabulary.
            transformer_width (int): The width of the transformer.
            transformer_heads (int): The number of attention heads in the transformer.
            transformer_layers (int): The number of layers in the transformer.
            use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
            rngs (rnglib.Rngs): The random number generator state. Defaults to nnx.Rngs(0).
            dtype (DTypeLike): The data type for computations. Defaults to jnp.float32.
            param_dtype (DTypeLike): The data type for parameters. Defaults to jnp.float32.
            mesh (Mesh | None): The device mesh for parameter sharding. Defaults to None.
        """
        self.vision_layers = vision_layers
        self.vision_width = vision_width
        self.vision_patch_size = vision_patch_size
        self.context_length = context_length
        self.vocab_size = vocab_size
        self.transformer_width = transformer_width
        self.transformer_heads = transformer_heads
        self.transformer_layers = transformer_layers
        self.dtype = dtype
        self._original_config = None

        self.attn_mask: Float[Array, "context_length context_length"] = jnp.tril(jnp.ones((context_length, context_length), dtype=dtype))

        self.vision_model = CLIPVisionModel(
            image_resolution=image_resolution,
            vision_layers=vision_layers,
            vision_width=vision_width,
            vision_patch_size=vision_patch_size,
            transformer_width=transformer_width,
            use_gradient_checkpointing=use_gradient_checkpointing,
            rngs=rngs,
            dtype=dtype,
            param_dtype=param_dtype,
            mesh=mesh,
        )

        self.text_model = Transformer(
            width=transformer_width,
            mlp_dim=transformer_width * 4,
            layers=transformer_layers,
            num_heads=transformer_heads,
            dropout_rate=0.0,
            attn_mask=self.attn_mask,
            use_quick_gelu=True,
            use_gradient_checkpointing=use_gradient_checkpointing,
            dtype=dtype,
            param_dtype=param_dtype,
            mesh=mesh,
            rngs=rngs,
        )
        self.vocab_size = vocab_size
        self.token_embedding = nnx.Embed(
            num_embeddings=vocab_size,
            features=transformer_width,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
            embedding_init=sharded_init(nnx.initializers.xavier_uniform(), P("model", None), mesh),
        )
        self.positional_embedding = nnx.Param(sharded_init(nnx.initializers.truncated_normal(stddev=0.02), P(None, "model"), mesh)(rngs.params(), (context_length, transformer_width)))
        self.text_position_ids = nnx.Param(jnp.arange(context_length, dtype=jnp.int32).reshape(1, -1))
        self.ln_final = nnx.LayerNorm(
            transformer_width,
            epsilon=1e-5,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
            scale_init=sharded_init(nnx.initializers.ones_init(), P("model"), mesh),
            bias_init=sharded_init(nnx.initializers.zeros_init(), P("model"), mesh),
        )
        self.text_projection = nnx.Linear(
            transformer_width,
            transformer_width,
            use_bias=False,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
            kernel_init=sharded_init(nnx.initializers.xavier_uniform(), P("model", None), mesh),
        )
        self.logit_scale = nnx.Param(sharded_init(nnx.initializers.ones_init(), P(), mesh)(rngs.params(), ()))

    def _create_config(self) -> dict[str, Any]:
        return {
            "model_type": "clip",
            "text_config": {
                "hidden_size": self.transformer_width,
                "num_attention_heads": self.transformer_heads,
                "num_hidden_layers": self.transformer_layers,
                "max_position_embeddings": self.context_length,
                "vocab_size": self.vocab_size,
            },
            "vision_config": {
                "hidden_size": self.vision_width,
                "num_attention_heads": self.vision_width // 64,
                "num_hidden_layers": self.vision_layers,
                "image_size": self.vision_model.encoder.img_size,
                "patch_size": self.vision_patch_size,
            },
        }

    def encode_image(self, image: Float[Array, "batch height width channels"], do_projection: bool = False) -> Float[Array, "batch transformer_width"]:
        """
        Encode images into embeddings.

        Args:
            image (Float[Array, "batch height width channels"]): Batch of input images.
            do_projection (bool): Whether the image encoder should do the visual projection layer.

        Returns:
            Float[Array, "batch transformer_width"]: Image embeddings.
        """
        return self.vision_model(image, do_projection)

    def encode_text(self, text: Int[Array, "batch context_length"]) -> Float[Array, "batch transformer_width"]:
        """
        Encode text tokens into embeddings.

        Args:
            text (Int[Array, "batch context_length"]): Batch of token sequences.

        Returns:
            Float[Array, "batch transformer_width"]: Text embeddings.
        """
        seq_len = text.shape[1]
        x: Float[Array, "batch context_length transformer_width"] = self.token_embedding(text)
        x: Float[Array, "batch context_length transformer_width"] = x + self.positional_embedding.value[:seq_len]
        x: Float[Array, "batch context_length transformer_width"] = self.text_model(x)
        x: Float[Array, "batch context_length transformer_width"] = self.ln_final(x)

        eot_token_pos: Float[Array, " batch "] = jnp.argmax(text, axis=-1)
        batch_indices: Float[Array, " batch "] = jnp.arange(x.shape[0])
        x: Float[Array, "batch transformer_width"] = x[batch_indices, eot_token_pos] @ self.text_projection.kernel.value
        return x

    def save_pretrained(self, save_directory: str) -> None:
        _SPECIAL_MAPPINGS = {
            "ln_final.weight": "text_model.final_layer_norm.weight",
            "ln_final.bias": "text_model.final_layer_norm.bias",
            "vision_model.encoder.ln_pre.weight": "vision_model.pre_layrnorm.weight",
            "vision_model.encoder.ln_pre.bias": "vision_model.pre_layrnorm.bias",
            "vision_model.encoder.ln_post.weight": "vision_model.post_layernorm.weight",
            "vision_model.encoder.ln_post.bias": "vision_model.post_layernorm.bias",
            "vision_model.encoder.cls_token": "vision_model.embeddings.class_embedding",
            "vision_model.encoder.position_embeddings": "vision_model.embeddings.position_embedding.weight",
            "vision_model.encoder.patch_embeddings.weight": "vision_model.embeddings.patch_embedding.weight",
            "positional_embedding": "text_model.embeddings.position_embedding.weight",
            "text_position_ids": "text_model.embeddings.position_ids",
            "vision_model.vision_position_ids": "vision_model.embeddings.position_ids",
            "token_embedding.embedding": "text_model.embeddings.token_embedding.weight",
            "text_projection.weight": "text_projection.weight",
            "vision_model.visual_projection.weight": "visual_projection.weight",
        }
        _SPECIAL_RENAMINGS = {
            "text_model.layers": "text_model.encoder.layers",
            "vision_model.encoder.encoder.layers": "vision_model.encoder.layers",
            ".attn.query.": ".self_attn.q_proj.",
            ".attn.key.": ".self_attn.k_proj.",
            ".attn.value.": ".self_attn.v_proj.",
            ".attn.out.": ".self_attn.out_proj.",
            ".mlp.layers.0.": ".mlp.fc1.",
            ".mlp.layers.3.": ".mlp.fc2.",
            ".norm1.": ".layer_norm1.",
            ".norm2.": ".layer_norm2.",
        }
        os.makedirs(save_directory, exist_ok=True)

        config = self._original_config.copy() if self._original_config else self._create_config()
        if jax.process_index() == 0:
            with open(os.path.join(save_directory, "config.json"), "w") as f:
                json.dump(config, f, indent=2)

        _, state = nnx.split(self)
        hf_state = convert_state_to_hf_format(nnx.to_pure_dict(state), _SPECIAL_MAPPINGS, _SPECIAL_RENAMINGS)
        save_safetensors(hf_state, os.path.join(save_directory, "model.safetensors"))

    def __call__(self, image: Float[Array, "batch height width channels"], text: Int[Array, "batch context_length"]) -> Float[Array, "batch batch"]:
        """
        Calculate similarity between image and text embeddings.

        Args:
            image (Float[Array, "batch height width channels"]): Batch of input images.
            text (Int[Array, "batch context_length"]): Batch of token sequences.

        Returns:
            Float[Array, "batch batch"]: Similarity scores between all pairs of images and texts.
        """
        image_features: Float[Array, "batch transformer_width"] = self.encode_image(image, do_projection=True)
        text_features: Float[Array, "batch transformer_width"] = self.encode_text(text)

        image_features: Float[Array, "batch transformer_width"] = image_features / jnp.linalg.norm(image_features, axis=-1, keepdims=True)
        text_features: Float[Array, "batch transformer_width"] = text_features / jnp.linalg.norm(text_features, axis=-1, keepdims=True)

        logit_scale: Float[Array, ""] = jnp.exp(self.logit_scale)
        logits: Float[Array, "batch batch"] = logit_scale * image_features @ text_features.T
        return logits

    @classmethod
    def from_pretrained(
        cls,
        model_name_or_path: str,
        use_pytorch: bool = False,
        mesh: Mesh | None = None,
        dtype: DTypeLike = jnp.float32,
        param_dtype: DTypeLike = jnp.float32,
        use_gradient_checkpointing: bool = False,
        rngs: rnglib.Rngs = nnx.Rngs(0),
    ) -> "CLIP":
        """Load a pretrained CLIP model from a local path or HuggingFace Hub.

        Args:
            model_name_or_path (str): Path to local weights or HuggingFace model ID.
            use_pytorch (bool): Whether to load from PyTorch weights. Defaults to False.
            mesh (Mesh | None): Optional device mesh for parameter sharding. Defaults to None.
            dtype (DTypeLike): Data type for computations. Defaults to jnp.float32.
            param_dtype (DTypeLike): Data type for parameters. Defaults to jnp.float32.
            use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
            rngs (rnglib.Rngs): Random number generator keys. Defaults to nnx.Rngs(0).

        Returns:
            CLIP: Pretrained CLIP model
        """

        params_fstate, config_dict = load_params_and_config(model_name_or_path, use_pytorch)

        config: dict[str, Any] = config_dict

        if config == {}:
            if not use_pytorch:
                text_hidden_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[1]
                text_max_pos_embed = params_fstate["text_model.embeddings.position_embedding.weight"].shape[0]
                text_vocab_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[0]

                text_num_layers = 0
                for k_param in params_fstate:
                    if k_param.startswith("text_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                        layer_idx = int(k_param.split(".")[3])
                        text_num_layers = max(text_num_layers, layer_idx + 1)

                vision_hidden_size = params_fstate["vision_model.embeddings.class_embedding"].shape[0]
                vision_patch_size = params_fstate["vision_model.embeddings.patch_embedding.weight"].shape[2]
                vision_image_size = int((params_fstate["vision_model.embeddings.position_embedding.weight"].shape[0] - 1) ** 0.5) * vision_patch_size

                vision_num_layers = 0
                for k_param in params_fstate:
                    if k_param.startswith("vision_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                        layer_idx = int(k_param.split(".")[3])
                        vision_num_layers = max(vision_num_layers, layer_idx + 1)

                config = {
                    "text_config": {
                        "hidden_size": text_hidden_size,
                        "num_attention_heads": text_hidden_size // 64,
                        "num_hidden_layers": text_num_layers,
                        "max_position_embeddings": text_max_pos_embed,
                        "vocab_size": text_vocab_size,
                    },
                    "vision_config": {
                        "hidden_size": vision_hidden_size,
                        "num_attention_heads": vision_hidden_size // 64,
                        "num_hidden_layers": vision_num_layers,
                        "image_size": vision_image_size,
                        "patch_size": vision_patch_size,
                    },
                }
            else:
                raise ValueError(f"Configuration could not be loaded for PyTorch model {model_name_or_path}")

        text_config = config["text_config"]
        vision_config = config["vision_config"]

        model = cls(
            image_resolution=vision_config["image_size"],
            vision_layers=vision_config["num_hidden_layers"],
            vision_width=vision_config["hidden_size"],
            vision_patch_size=vision_config["patch_size"],
            context_length=text_config["max_position_embeddings"],
            vocab_size=text_config["vocab_size"],
            transformer_width=text_config["hidden_size"],
            transformer_heads=text_config["num_attention_heads"],
            transformer_layers=text_config["num_hidden_layers"],
            use_gradient_checkpointing=use_gradient_checkpointing,
            mesh=mesh,
            dtype=dtype,
            param_dtype=param_dtype,
            rngs=rngs,
        )

        flax_model_params_fstate = dict(nnx.to_flat_state(nnx.state(model, nnx.Param)))

        mapping_list = [
            (("logit_scale",), ("logit_scale",)),
            (("positional_embedding",), ("text_model", "embeddings", "position_embedding", "weight")),
            (("text_position_ids",), ("text_model", "embeddings", "position_ids")),
            (("token_embedding", "embedding"), ("text_model", "embeddings", "token_embedding", "weight")),
            (("ln_final", "scale"), ("text_model", "final_layer_norm", "weight")),
            (("ln_final", "bias"), ("text_model", "final_layer_norm", "bias")),
            (("text_projection", "kernel"), ("text_projection", "weight")),
            (("vision_model", "encoder", "cls_token"), ("vision_model", "embeddings", "class_embedding")),
            (("vision_model", "encoder", "position_embeddings"), ("vision_model", "embeddings", "position_embedding", "weight")),
            (("vision_model", "encoder", "patch_embeddings", "kernel"), ("vision_model", "embeddings", "patch_embedding", "weight")),
            (("vision_model", "encoder", "ln_pre", "scale"), ("vision_model", "pre_layrnorm", "weight")),
            (("vision_model", "encoder", "ln_pre", "bias"), ("vision_model", "pre_layrnorm", "bias")),
            (("vision_model", "encoder", "ln_post", "scale"), ("vision_model", "post_layernorm", "weight")),
            (("vision_model", "encoder", "ln_post", "bias"), ("vision_model", "post_layernorm", "bias")),
            (("vision_model", "visual_projection", "kernel"), ("visual_projection", "weight")),
        ]

        for i in range(text_config["num_hidden_layers"]):
            flax_base = ("text_model", "layers", i)
            hf_base = ("text_model", "encoder", "layers", str(i))

            mapping_list.extend(
                [
                    (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                    (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                    (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                    (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                    (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                    (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                    (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                    (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                    (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                    (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                    (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                    (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                    (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                    (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                    (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                    (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
                ]
            )

        for i in range(vision_config["num_hidden_layers"]):
            flax_base = ("vision_model", "encoder", "encoder", "layers", i)
            hf_base = ("vision_model", "encoder", "layers", str(i))

            mapping_list.extend(
                [
                    (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                    (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                    (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                    (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                    (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                    (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                    (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                    (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                    (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                    (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                    (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                    (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                    (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                    (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                    (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                    (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
                ]
            )

        params_name_mapping = dict(mapping_list)
        nonvisited = set(flax_model_params_fstate.keys())

        hf_checkpoint_keys: Set[str] = set(params_fstate.keys())
        used_hf_keys: Set[str] = set()

        for flax_dst_key_tuple, hf_src_key_tuple in params_name_mapping.items():
            hf_src_key_as_string = ".".join(hf_src_key_tuple)

            used_hf_keys.add(hf_src_key_as_string)
            nonvisited.discard(flax_dst_key_tuple)
            src_value = params_fstate[hf_src_key_as_string]
            dst_value_obj = flax_model_params_fstate[flax_dst_key_tuple]

            if flax_dst_key_tuple[0] == "vision_model":
                if flax_dst_key_tuple == ("vision_model", "encoder", "patch_embeddings", "kernel"):
                    src_value = jnp.transpose(src_value, (2, 3, 1, 0))
                elif flax_dst_key_tuple == ("vision_model", "encoder", "cls_token"):
                    src_value = src_value.reshape(1, 1, -1)
                elif flax_dst_key_tuple == ("vision_model", "encoder", "position_embeddings"):
                    src_value = src_value.reshape(1, src_value.shape[0], src_value.shape[1])
                elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                    src_value = jnp.transpose(src_value, (1, 0))
                    num_heads = vision_config["hidden_size"] // 64
                    hidden_size = vision_config["hidden_size"]
                    head_dim = hidden_size // num_heads
                    src_value = src_value.reshape((hidden_size, num_heads, head_dim))
                elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                    num_heads = vision_config["hidden_size"] // 64
                    hidden_size = vision_config["hidden_size"]
                    head_dim = hidden_size // num_heads
                    src_value = src_value.reshape((num_heads, head_dim))
                elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
                    src_value = jnp.transpose(src_value, (1, 0))
                    num_heads = vision_config["hidden_size"] // 64
                    hidden_size = vision_config["hidden_size"]
                    head_dim = hidden_size // num_heads
                    src_value = src_value.reshape((num_heads, head_dim, hidden_size))
                elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
                    src_value = jnp.transpose(src_value, (1, 0))
            elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = text_config["num_attention_heads"]
                hidden_size = text_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((hidden_size, num_heads, head_dim))
            elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                num_heads = text_config["num_attention_heads"]
                hidden_size = text_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim))
            elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = text_config["num_attention_heads"]
                hidden_size = text_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim, hidden_size))
            elif flax_dst_key_tuple == ("token_embedding", "embedding") or flax_dst_key_tuple == ("positional_embedding",):
                pass
            elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
                src_value = jnp.transpose(src_value, (1, 0))

            if src_value.shape != dst_value_obj.value.shape:
                raise ValueError(f"Shape mismatch for {flax_dst_key_tuple} (Flax) vs {hf_src_key_as_string} (HF): {dst_value_obj.value.shape} (expected) != {src_value.shape} (actual)")

            src_value = src_value.astype(param_dtype)
            dst_value_obj.value = src_value

        nnx.update(model, nnx.from_flat_state(flax_model_params_fstate))

        known_buffer_keys = {
            ("vision_model", "encoder", "vision_position_ids"),
        }
        unexpected_nonvisited = nonvisited - known_buffer_keys
        if unexpected_nonvisited:
            print(f"Warning: Some CLIP parameters were not loaded: {sorted(list(unexpected_nonvisited))}")

        leftover_hf_keys = hf_checkpoint_keys - used_hf_keys

        known_unused_hf_buffer_keys = {
            "text_model.embeddings.position_ids",
            "vision_model.embeddings.position_ids",
            "vision_model.encoder.vision_position_ids",
        }
        unexpected_leftover_hf_keys = leftover_hf_keys - known_unused_hf_buffer_keys

        assert len(unexpected_leftover_hf_keys) == 0, f"Some unexpected HuggingFace checkpoint parameters were not used: {sorted(list(unexpected_leftover_hf_keys))}"

        model._original_config = config
        return model

__call__(image, text)

Calculate similarity between image and text embeddings.

Parameters:

Name Type Description Default
image Float[Array, 'batch height width channels']

Batch of input images.

 required
text Int[Array, 'batch context_length']

Batch of token sequences.

 required

Returns:

Type Description
Float[Array, 'batch batch']

Float[Array, "batch batch"]: Similarity scores between all pairs of images and texts.
Source code in src/jimm/models/clip.py 
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
def __call__(self, image: Float[Array, "batch height width channels"], text: Int[Array, "batch context_length"]) -> Float[Array, "batch batch"]:
    """
    Calculate similarity between image and text embeddings.

    Args:
        image (Float[Array, "batch height width channels"]): Batch of input images.
        text (Int[Array, "batch context_length"]): Batch of token sequences.

    Returns:
        Float[Array, "batch batch"]: Similarity scores between all pairs of images and texts.
    """
    image_features: Float[Array, "batch transformer_width"] = self.encode_image(image, do_projection=True)
    text_features: Float[Array, "batch transformer_width"] = self.encode_text(text)

    image_features: Float[Array, "batch transformer_width"] = image_features / jnp.linalg.norm(image_features, axis=-1, keepdims=True)
    text_features: Float[Array, "batch transformer_width"] = text_features / jnp.linalg.norm(text_features, axis=-1, keepdims=True)

    logit_scale: Float[Array, ""] = jnp.exp(self.logit_scale)
    logits: Float[Array, "batch batch"] = logit_scale * image_features @ text_features.T
    return logits

__init__(image_resolution, vision_layers, vision_width, vision_patch_size, context_length, vocab_size, transformer_width, transformer_heads, transformer_layers, use_gradient_checkpointing=False, rngs=nnx.Rngs(0), dtype=jnp.float32, param_dtype=jnp.float32, mesh=None)

Initialize the CLIP model.

Parameters:

Name Type Description Default
image_resolution int

The resolution of the input images.

 required
vision_layers int

The number of layers in the vision transformer.

 required
vision_width int

The width of the vision transformer.

 required
vision_patch_size int

The patch size of the vision transformer.

 required
context_length int

The length of the context.

 required
vocab_size int

The size of the vocabulary.

 required
transformer_width int

The width of the transformer.

 required
transformer_heads int

The number of attention heads in the transformer.

 required
transformer_layers int

The number of layers in the transformer.

 required
use_gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
rngs Rngs

The random number generator state. Defaults to nnx.Rngs(0).

 Rngs(0)
dtype DTypeLike

The data type for computations. Defaults to jnp.float32.

 float32
param_dtype DTypeLike

The data type for parameters. Defaults to jnp.float32.

 float32
mesh Mesh | None

The device mesh for parameter sharding. Defaults to None.

 None
Source code in src/jimm/models/clip.py 
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
def __init__(
    self,
    image_resolution: int,
    vision_layers: int,
    vision_width: int,
    vision_patch_size: int,
    context_length: int,
    vocab_size: int,
    transformer_width: int,
    transformer_heads: int,
    transformer_layers: int,
    use_gradient_checkpointing: bool = False,
    rngs: rnglib.Rngs = nnx.Rngs(0),
    dtype: DTypeLike = jnp.float32,
    param_dtype: DTypeLike = jnp.float32,
    mesh: Mesh | None = None,
):
    """
    Initialize the CLIP model.

    Args:
        image_resolution (int): The resolution of the input images.
        vision_layers (int): The number of layers in the vision transformer.
        vision_width (int): The width of the vision transformer.
        vision_patch_size (int): The patch size of the vision transformer.
        context_length (int): The length of the context.
        vocab_size (int): The size of the vocabulary.
        transformer_width (int): The width of the transformer.
        transformer_heads (int): The number of attention heads in the transformer.
        transformer_layers (int): The number of layers in the transformer.
        use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
        rngs (rnglib.Rngs): The random number generator state. Defaults to nnx.Rngs(0).
        dtype (DTypeLike): The data type for computations. Defaults to jnp.float32.
        param_dtype (DTypeLike): The data type for parameters. Defaults to jnp.float32.
        mesh (Mesh | None): The device mesh for parameter sharding. Defaults to None.
    """
    self.vision_layers = vision_layers
    self.vision_width = vision_width
    self.vision_patch_size = vision_patch_size
    self.context_length = context_length
    self.vocab_size = vocab_size
    self.transformer_width = transformer_width
    self.transformer_heads = transformer_heads
    self.transformer_layers = transformer_layers
    self.dtype = dtype
    self._original_config = None

    self.attn_mask: Float[Array, "context_length context_length"] = jnp.tril(jnp.ones((context_length, context_length), dtype=dtype))

    self.vision_model = CLIPVisionModel(
        image_resolution=image_resolution,
        vision_layers=vision_layers,
        vision_width=vision_width,
        vision_patch_size=vision_patch_size,
        transformer_width=transformer_width,
        use_gradient_checkpointing=use_gradient_checkpointing,
        rngs=rngs,
        dtype=dtype,
        param_dtype=param_dtype,
        mesh=mesh,
    )

    self.text_model = Transformer(
        width=transformer_width,
        mlp_dim=transformer_width * 4,
        layers=transformer_layers,
        num_heads=transformer_heads,
        dropout_rate=0.0,
        attn_mask=self.attn_mask,
        use_quick_gelu=True,
        use_gradient_checkpointing=use_gradient_checkpointing,
        dtype=dtype,
        param_dtype=param_dtype,
        mesh=mesh,
        rngs=rngs,
    )
    self.vocab_size = vocab_size
    self.token_embedding = nnx.Embed(
        num_embeddings=vocab_size,
        features=transformer_width,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
        embedding_init=sharded_init(nnx.initializers.xavier_uniform(), P("model", None), mesh),
    )
    self.positional_embedding = nnx.Param(sharded_init(nnx.initializers.truncated_normal(stddev=0.02), P(None, "model"), mesh)(rngs.params(), (context_length, transformer_width)))
    self.text_position_ids = nnx.Param(jnp.arange(context_length, dtype=jnp.int32).reshape(1, -1))
    self.ln_final = nnx.LayerNorm(
        transformer_width,
        epsilon=1e-5,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
        scale_init=sharded_init(nnx.initializers.ones_init(), P("model"), mesh),
        bias_init=sharded_init(nnx.initializers.zeros_init(), P("model"), mesh),
    )
    self.text_projection = nnx.Linear(
        transformer_width,
        transformer_width,
        use_bias=False,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
        kernel_init=sharded_init(nnx.initializers.xavier_uniform(), P("model", None), mesh),
    )
    self.logit_scale = nnx.Param(sharded_init(nnx.initializers.ones_init(), P(), mesh)(rngs.params(), ()))

encode_image(image, do_projection=False)

Encode images into embeddings.

Parameters:

Name Type Description Default
image Float[Array, 'batch height width channels']

Batch of input images.

 required
do_projection bool

Whether the image encoder should do the visual projection layer.

 False

Returns:

Type Description
Float[Array, 'batch transformer_width']

Float[Array, "batch transformer_width"]: Image embeddings.
Source code in src/jimm/models/clip.py 
399
400
401
402
403
404
405
406
407
408
409
410
def encode_image(self, image: Float[Array, "batch height width channels"], do_projection: bool = False) -> Float[Array, "batch transformer_width"]:
    """
    Encode images into embeddings.

    Args:
        image (Float[Array, "batch height width channels"]): Batch of input images.
        do_projection (bool): Whether the image encoder should do the visual projection layer.

    Returns:
        Float[Array, "batch transformer_width"]: Image embeddings.
    """
    return self.vision_model(image, do_projection)

encode_text(text)

Encode text tokens into embeddings.

Parameters:

Name Type Description Default
text Int[Array, 'batch context_length']

Batch of token sequences.

 required

Returns:

Type Description
Float[Array, 'batch transformer_width']

Float[Array, "batch transformer_width"]: Text embeddings.
Source code in src/jimm/models/clip.py 
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
def encode_text(self, text: Int[Array, "batch context_length"]) -> Float[Array, "batch transformer_width"]:
    """
    Encode text tokens into embeddings.

    Args:
        text (Int[Array, "batch context_length"]): Batch of token sequences.

    Returns:
        Float[Array, "batch transformer_width"]: Text embeddings.
    """
    seq_len = text.shape[1]
    x: Float[Array, "batch context_length transformer_width"] = self.token_embedding(text)
    x: Float[Array, "batch context_length transformer_width"] = x + self.positional_embedding.value[:seq_len]
    x: Float[Array, "batch context_length transformer_width"] = self.text_model(x)
    x: Float[Array, "batch context_length transformer_width"] = self.ln_final(x)

    eot_token_pos: Float[Array, " batch "] = jnp.argmax(text, axis=-1)
    batch_indices: Float[Array, " batch "] = jnp.arange(x.shape[0])
    x: Float[Array, "batch transformer_width"] = x[batch_indices, eot_token_pos] @ self.text_projection.kernel.value
    return x

from_pretrained(model_name_or_path, use_pytorch=False, mesh=None, dtype=jnp.float32, param_dtype=jnp.float32, use_gradient_checkpointing=False, rngs=nnx.Rngs(0)) classmethod

Load a pretrained CLIP model from a local path or HuggingFace Hub.

Parameters:

Name Type Description Default
model_name_or_path str

Path to local weights or HuggingFace model ID.

 required
use_pytorch bool

Whether to load from PyTorch weights. Defaults to False.

 False
mesh Mesh | None

Optional device mesh for parameter sharding. Defaults to None.

 None
dtype DTypeLike

Data type for computations. Defaults to jnp.float32.

 float32
param_dtype DTypeLike

Data type for parameters. Defaults to jnp.float32.

 float32
use_gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
rngs Rngs

Random number generator keys. Defaults to nnx.Rngs(0).

 Rngs(0)

Returns:

Name Type Description
CLIP CLIP

Pretrained CLIP model
Source code in src/jimm/models/clip.py 
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
@classmethod
def from_pretrained(
    cls,
    model_name_or_path: str,
    use_pytorch: bool = False,
    mesh: Mesh | None = None,
    dtype: DTypeLike = jnp.float32,
    param_dtype: DTypeLike = jnp.float32,
    use_gradient_checkpointing: bool = False,
    rngs: rnglib.Rngs = nnx.Rngs(0),
) -> "CLIP":
    """Load a pretrained CLIP model from a local path or HuggingFace Hub.

    Args:
        model_name_or_path (str): Path to local weights or HuggingFace model ID.
        use_pytorch (bool): Whether to load from PyTorch weights. Defaults to False.
        mesh (Mesh | None): Optional device mesh for parameter sharding. Defaults to None.
        dtype (DTypeLike): Data type for computations. Defaults to jnp.float32.
        param_dtype (DTypeLike): Data type for parameters. Defaults to jnp.float32.
        use_gradient_checkpointing (bool): Whether to use gradient checkpointing. Defaults to False.
        rngs (rnglib.Rngs): Random number generator keys. Defaults to nnx.Rngs(0).

    Returns:
        CLIP: Pretrained CLIP model
    """

    params_fstate, config_dict = load_params_and_config(model_name_or_path, use_pytorch)

    config: dict[str, Any] = config_dict

    if config == {}:
        if not use_pytorch:
            text_hidden_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[1]
            text_max_pos_embed = params_fstate["text_model.embeddings.position_embedding.weight"].shape[0]
            text_vocab_size = params_fstate["text_model.embeddings.token_embedding.weight"].shape[0]

            text_num_layers = 0
            for k_param in params_fstate:
                if k_param.startswith("text_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                    layer_idx = int(k_param.split(".")[3])
                    text_num_layers = max(text_num_layers, layer_idx + 1)

            vision_hidden_size = params_fstate["vision_model.embeddings.class_embedding"].shape[0]
            vision_patch_size = params_fstate["vision_model.embeddings.patch_embedding.weight"].shape[2]
            vision_image_size = int((params_fstate["vision_model.embeddings.position_embedding.weight"].shape[0] - 1) ** 0.5) * vision_patch_size

            vision_num_layers = 0
            for k_param in params_fstate:
                if k_param.startswith("vision_model.encoder.layers.") and k_param.endswith(".self_attn.q_proj.weight"):
                    layer_idx = int(k_param.split(".")[3])
                    vision_num_layers = max(vision_num_layers, layer_idx + 1)

            config = {
                "text_config": {
                    "hidden_size": text_hidden_size,
                    "num_attention_heads": text_hidden_size // 64,
                    "num_hidden_layers": text_num_layers,
                    "max_position_embeddings": text_max_pos_embed,
                    "vocab_size": text_vocab_size,
                },
                "vision_config": {
                    "hidden_size": vision_hidden_size,
                    "num_attention_heads": vision_hidden_size // 64,
                    "num_hidden_layers": vision_num_layers,
                    "image_size": vision_image_size,
                    "patch_size": vision_patch_size,
                },
            }
        else:
            raise ValueError(f"Configuration could not be loaded for PyTorch model {model_name_or_path}")

    text_config = config["text_config"]
    vision_config = config["vision_config"]

    model = cls(
        image_resolution=vision_config["image_size"],
        vision_layers=vision_config["num_hidden_layers"],
        vision_width=vision_config["hidden_size"],
        vision_patch_size=vision_config["patch_size"],
        context_length=text_config["max_position_embeddings"],
        vocab_size=text_config["vocab_size"],
        transformer_width=text_config["hidden_size"],
        transformer_heads=text_config["num_attention_heads"],
        transformer_layers=text_config["num_hidden_layers"],
        use_gradient_checkpointing=use_gradient_checkpointing,
        mesh=mesh,
        dtype=dtype,
        param_dtype=param_dtype,
        rngs=rngs,
    )

    flax_model_params_fstate = dict(nnx.to_flat_state(nnx.state(model, nnx.Param)))

    mapping_list = [
        (("logit_scale",), ("logit_scale",)),
        (("positional_embedding",), ("text_model", "embeddings", "position_embedding", "weight")),
        (("text_position_ids",), ("text_model", "embeddings", "position_ids")),
        (("token_embedding", "embedding"), ("text_model", "embeddings", "token_embedding", "weight")),
        (("ln_final", "scale"), ("text_model", "final_layer_norm", "weight")),
        (("ln_final", "bias"), ("text_model", "final_layer_norm", "bias")),
        (("text_projection", "kernel"), ("text_projection", "weight")),
        (("vision_model", "encoder", "cls_token"), ("vision_model", "embeddings", "class_embedding")),
        (("vision_model", "encoder", "position_embeddings"), ("vision_model", "embeddings", "position_embedding", "weight")),
        (("vision_model", "encoder", "patch_embeddings", "kernel"), ("vision_model", "embeddings", "patch_embedding", "weight")),
        (("vision_model", "encoder", "ln_pre", "scale"), ("vision_model", "pre_layrnorm", "weight")),
        (("vision_model", "encoder", "ln_pre", "bias"), ("vision_model", "pre_layrnorm", "bias")),
        (("vision_model", "encoder", "ln_post", "scale"), ("vision_model", "post_layernorm", "weight")),
        (("vision_model", "encoder", "ln_post", "bias"), ("vision_model", "post_layernorm", "bias")),
        (("vision_model", "visual_projection", "kernel"), ("visual_projection", "weight")),
    ]

    for i in range(text_config["num_hidden_layers"]):
        flax_base = ("text_model", "layers", i)
        hf_base = ("text_model", "encoder", "layers", str(i))

        mapping_list.extend(
            [
                (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
            ]
        )

    for i in range(vision_config["num_hidden_layers"]):
        flax_base = ("vision_model", "encoder", "encoder", "layers", i)
        hf_base = ("vision_model", "encoder", "layers", str(i))

        mapping_list.extend(
            [
                (flax_base + ("attn", "query", "kernel"), hf_base + ("self_attn", "q_proj", "weight")),
                (flax_base + ("attn", "query", "bias"), hf_base + ("self_attn", "q_proj", "bias")),
                (flax_base + ("attn", "key", "kernel"), hf_base + ("self_attn", "k_proj", "weight")),
                (flax_base + ("attn", "key", "bias"), hf_base + ("self_attn", "k_proj", "bias")),
                (flax_base + ("attn", "value", "kernel"), hf_base + ("self_attn", "v_proj", "weight")),
                (flax_base + ("attn", "value", "bias"), hf_base + ("self_attn", "v_proj", "bias")),
                (flax_base + ("attn", "out", "kernel"), hf_base + ("self_attn", "out_proj", "weight")),
                (flax_base + ("attn", "out", "bias"), hf_base + ("self_attn", "out_proj", "bias")),
                (flax_base + ("norm1", "scale"), hf_base + ("layer_norm1", "weight")),
                (flax_base + ("norm1", "bias"), hf_base + ("layer_norm1", "bias")),
                (flax_base + ("norm2", "scale"), hf_base + ("layer_norm2", "weight")),
                (flax_base + ("norm2", "bias"), hf_base + ("layer_norm2", "bias")),
                (flax_base + ("mlp", "layers", 0, "kernel"), hf_base + ("mlp", "fc1", "weight")),
                (flax_base + ("mlp", "layers", 0, "bias"), hf_base + ("mlp", "fc1", "bias")),
                (flax_base + ("mlp", "layers", 3, "kernel"), hf_base + ("mlp", "fc2", "weight")),
                (flax_base + ("mlp", "layers", 3, "bias"), hf_base + ("mlp", "fc2", "bias")),
            ]
        )

    params_name_mapping = dict(mapping_list)
    nonvisited = set(flax_model_params_fstate.keys())

    hf_checkpoint_keys: Set[str] = set(params_fstate.keys())
    used_hf_keys: Set[str] = set()

    for flax_dst_key_tuple, hf_src_key_tuple in params_name_mapping.items():
        hf_src_key_as_string = ".".join(hf_src_key_tuple)

        used_hf_keys.add(hf_src_key_as_string)
        nonvisited.discard(flax_dst_key_tuple)
        src_value = params_fstate[hf_src_key_as_string]
        dst_value_obj = flax_model_params_fstate[flax_dst_key_tuple]

        if flax_dst_key_tuple[0] == "vision_model":
            if flax_dst_key_tuple == ("vision_model", "encoder", "patch_embeddings", "kernel"):
                src_value = jnp.transpose(src_value, (2, 3, 1, 0))
            elif flax_dst_key_tuple == ("vision_model", "encoder", "cls_token"):
                src_value = src_value.reshape(1, 1, -1)
            elif flax_dst_key_tuple == ("vision_model", "encoder", "position_embeddings"):
                src_value = src_value.reshape(1, src_value.shape[0], src_value.shape[1])
            elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((hidden_size, num_heads, head_dim))
            elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim))
            elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
                src_value = jnp.transpose(src_value, (1, 0))
                num_heads = vision_config["hidden_size"] // 64
                hidden_size = vision_config["hidden_size"]
                head_dim = hidden_size // num_heads
                src_value = src_value.reshape((num_heads, head_dim, hidden_size))
            elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
                src_value = jnp.transpose(src_value, (1, 0))
        elif hf_src_key_tuple[-1] == "weight" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
            src_value = jnp.transpose(src_value, (1, 0))
            num_heads = text_config["num_attention_heads"]
            hidden_size = text_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((hidden_size, num_heads, head_dim))
        elif hf_src_key_tuple[-1] == "bias" and hf_src_key_tuple[-2] in ("q_proj", "k_proj", "v_proj"):
            num_heads = text_config["num_attention_heads"]
            hidden_size = text_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((num_heads, head_dim))
        elif hf_src_key_tuple[-2:] == ("out_proj", "weight"):
            src_value = jnp.transpose(src_value, (1, 0))
            num_heads = text_config["num_attention_heads"]
            hidden_size = text_config["hidden_size"]
            head_dim = hidden_size // num_heads
            src_value = src_value.reshape((num_heads, head_dim, hidden_size))
        elif flax_dst_key_tuple == ("token_embedding", "embedding") or flax_dst_key_tuple == ("positional_embedding",):
            pass
        elif hf_src_key_tuple[-1] == "weight" and src_value.ndim == 2:
            src_value = jnp.transpose(src_value, (1, 0))

        if src_value.shape != dst_value_obj.value.shape:
            raise ValueError(f"Shape mismatch for {flax_dst_key_tuple} (Flax) vs {hf_src_key_as_string} (HF): {dst_value_obj.value.shape} (expected) != {src_value.shape} (actual)")

        src_value = src_value.astype(param_dtype)
        dst_value_obj.value = src_value

    nnx.update(model, nnx.from_flat_state(flax_model_params_fstate))

    known_buffer_keys = {
        ("vision_model", "encoder", "vision_position_ids"),
    }
    unexpected_nonvisited = nonvisited - known_buffer_keys
    if unexpected_nonvisited:
        print(f"Warning: Some CLIP parameters were not loaded: {sorted(list(unexpected_nonvisited))}")

    leftover_hf_keys = hf_checkpoint_keys - used_hf_keys

    known_unused_hf_buffer_keys = {
        "text_model.embeddings.position_ids",
        "vision_model.embeddings.position_ids",
        "vision_model.encoder.vision_position_ids",
    }
    unexpected_leftover_hf_keys = leftover_hf_keys - known_unused_hf_buffer_keys

    assert len(unexpected_leftover_hf_keys) == 0, f"Some unexpected HuggingFace checkpoint parameters were not used: {sorted(list(unexpected_leftover_hf_keys))}"

    model._original_config = config
    return model