Skip to content

Models

InputHandler

Source code in easyroutine/interpretability/models.py 
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
class InputHandler:
    def __init__(self, model_name: str):
        self.model_name = model_name

    def prepare_inputs(
        self,
        batch_dict: Dict[str, torch.Tensor],
        device: Union[str, torch.device],
        torch_dtype: torch.dtype = torch.bfloat16,
        require_grads: bool = False,
    ):
        if self.model_name in [
            "facebook/chameleon-7b",
            "facebook/chameleon-30b",
        ]:
            input_dict = {
                "input_ids": batch_dict["input_ids"],
                "attention_mask": batch_dict["attention_mask"],
                "pixel_values": batch_dict["pixel_values"].to(torch_dtype),
            }
        elif self.model_name in ["mistral-community/pixtral-12b"]:
            if "pixel_values" not in batch_dict:
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                }
            else:
                if (
                    isinstance(batch_dict["pixel_values"][0], list)
                    and len(batch_dict["pixel_values"]) == 1
                ):
                    # batch_dict["pixel_values"] = batch_dict["pixel_values"][0]
                    batch_dict["pixel_values"] = [
                        [
                            image.to(torch_dtype)
                            for image in batch_dict["pixel_values"][0]
                        ]
                    ]
                elif isinstance(batch_dict["pixel_values"], torch.Tensor):
                    batch_dict["pixel_values"] = batch_dict["pixel_values"].to(
                        torch_dtype
                    )
                elif isinstance(batch_dict["pixel_values"], list):
                    batch_dict["pixel_values"] = [
                        image.to(torch_dtype) for image in batch_dict["pixel_values"]
                    ]
                else:
                    raise ValueError("Pixel values not recognized. Please fix!")
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                    "pixel_values": batch_dict["pixel_values"],
                }

        elif self.model_name in ["meta-llama/Llama-3.2-1B", "meta-llama/Llama-3.2-3B"]:
            input_dict = {
                "input_ids": batch_dict["input_ids"],
                "attention_mask": batch_dict["attention_mask"],
            }
        elif self.model_name in ["Emu3-Chat", "Emu3-Gen", "Emu3-Stage1"]:
            raise NotImplementedError("Emu3 model not implemented yet")
        elif self.model_name in ["hf-internal-testing/tiny-random-LlamaForCausalLM"]:
            input_dict = {
                "input_ids": batch_dict["input_ids"],
                "attention_mask": batch_dict["attention_mask"],
            }
        elif self.model_name in ["llava-hf/llava-v1.6-mistral-7b-hf"]:
            if "pixel_values" not in batch_dict:
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                }
            else:
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                    "pixel_values": batch_dict["pixel_values"],
                    "image_sizes": batch_dict["image_sizes"],
                }
        elif self.model_name in ["google/gemma-3-1b-it", "google/gemma-3-4b-it", "google/gemma-3-12b-it", "google/gemma-3-27b-it"]:
            if "pixel_values" not in batch_dict:
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                }
            else:
                input_dict = {
                    "input_ids": batch_dict["input_ids"],
                    "attention_mask": batch_dict["attention_mask"],
                    "pixel_values": batch_dict["pixel_values"],
                    "token_type_ids": batch_dict["token_type_ids"],
                }
        elif self.model_name in ["CohereForAI/aya-101"]:
            input_dict = {
                "input_ids": batch_dict["input_ids"],
                "decoder_input_ids": batch_dict["input_ids"],
                "attention_mask": batch_dict["attention_mask"],
            }
        else:
            raise ValueError(f"Unsupported model_name: {self.model_name}")

        for key, value in input_dict.items():
            if isinstance(value, torch.Tensor):
                value = value.to(device)
            elif isinstance(value, list):
                if isinstance(value[0], torch.Tensor):
                    value = [v.to(device) for v in value]
                elif isinstance(value[0], list):
                    value = [[v.to(device) for v in vv] for vv in value]
                else:
                    raise ValueError(
                        f"Problem while moving the input to the device. The input with key {key} is not a torch.Tensor, a list of torch.Tensor or a list of list of torch.Tensor."
                    )

            input_dict[key] = value

        if require_grads:
            input_dict["input_ids"].requires_grad = True
        return input_dict

    def get_input_ids(
        self,
        input_dict: Dict[str, torch.Tensor],
    ):
        return input_dict["input_ids"]

    def cleanup_tensors(self, inputs: Dict[str, Any], others: Dict[str, Any] = None):
        """
        Clean up tensors to free GPU memory by detaching, moving to CPU, and deleting.

        Args:
            inputs (Dict[str, Any]): Dictionary of input tensors (usually from prepare_inputs)
            others (Dict[str, Any], optional): Dictionary of other tensors not in inputs

        Returns:
            None
        """
        # Clean up inputs from the batch
        if inputs is not None:
            for k_in in list(inputs.keys()):  # Iterate over a copy of keys
                if isinstance(inputs[k_in], torch.Tensor):
                    inputs[k_in] = inputs[k_in].detach().cpu()
                del inputs[k_in]
            del inputs

        # Clean up others from the batch
        if others is not None:
            for k_oth in list(others.keys()):
                if isinstance(others[k_oth], torch.Tensor):
                    others[k_oth] = others[k_oth].detach().cpu()
                del others[k_oth]
            del others

        # Explicitly run garbage collection to free memory
        torch.cuda.empty_cache()

cleanup_tensors(inputs, others=None)

Clean up tensors to free GPU memory by detaching, moving to CPU, and deleting.

Parameters:

Name Type Description Default
inputs Dict[str, Any]

Dictionary of input tensors (usually from prepare_inputs)

 required
others Dict[str, Any]

Dictionary of other tensors not in inputs

 None

Returns:

Type Description

None
Source code in easyroutine/interpretability/models.py 
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
def cleanup_tensors(self, inputs: Dict[str, Any], others: Dict[str, Any] = None):
    """
    Clean up tensors to free GPU memory by detaching, moving to CPU, and deleting.

    Args:
        inputs (Dict[str, Any]): Dictionary of input tensors (usually from prepare_inputs)
        others (Dict[str, Any], optional): Dictionary of other tensors not in inputs

    Returns:
        None
    """
    # Clean up inputs from the batch
    if inputs is not None:
        for k_in in list(inputs.keys()):  # Iterate over a copy of keys
            if isinstance(inputs[k_in], torch.Tensor):
                inputs[k_in] = inputs[k_in].detach().cpu()
            del inputs[k_in]
        del inputs

    # Clean up others from the batch
    if others is not None:
        for k_oth in list(others.keys()):
            if isinstance(others[k_oth], torch.Tensor):
                others[k_oth] = others[k_oth].detach().cpu()
            del others[k_oth]
        del others

    # Explicitly run garbage collection to free memory
    torch.cuda.empty_cache()

ModelConfig dataclass

Configuration class for storing model specific parameters.

Attributes:

Name Type Description
residual_stream_input_hook_name str

Name of the residual stream torch module where attach the hook
residual_stream_hook_name str

Name of the residual stram torch module where attach the hook
intermediate_stream_hook_name str

Name of the intermediate stream torch module where attach the hook
residual_stream_input_post_layernorm_hook_name str

Name of the residual stream input post layer norm
head_key_hook_name str

Name of the attention key torch module where attach the hook
head_value_hook_name str

Name of the attention value torch module where attach the hook
head_query_hook_name str

Name of the attention key torch module where attach the hook
attn_in_hook_name str

Name of the attention input torch module where attach the hook
attn_out_hook_name str

Name of the attention output torch module where attach the hook
attn_matrix_hook_name str

Name of the attention matrix torch module where attach the hook
mlp_out_hook_name str

Name of the mlp output torch module where attach the hook
attn_out_proj_weight str

Name of the attention output projection weight
attn_out_proj_bias str

Name of the attention output projection bias
embed_tokens str

Name of the embedding tokens torch module where attach the hook
num_hidden_layers int

Number of hidden layers
num_attention_heads int

Number of attention heads
hidden_size int

Hidden size of the transformer model
num_key_value_heads int

Number of key value heads
num_key_value_groups int

Number of key value groups
head_dim int

Dimension of the attention head
Source code in easyroutine/interpretability/models.py 
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
@dataclass
class ModelConfig:
    r"""
    Configuration class for storing model specific parameters.

    Attributes:
        residual_stream_input_hook_name (str): Name of the residual stream torch module where attach the hook
        residual_stream_hook_name (str): Name of the residual stram torch module where attach the hook
        intermediate_stream_hook_name (str): Name of the intermediate stream torch module where attach the hook
        residual_stream_input_post_layernorm_hook_name (str): Name of the residual stream input post layer norm
        head_key_hook_name (str): Name of the attention key torch module where attach the hook
        head_value_hook_name (str): Name of the attention value torch module where attach the hook
        head_query_hook_name (str): Name of the attention key torch module where attach the hook
        attn_in_hook_name (str): Name of the attention input torch module where attach the hook
        attn_out_hook_name (str): Name of the attention output torch module where attach the hook
        attn_matrix_hook_name (str): Name of the attention matrix torch module where attach the hook
        mlp_out_hook_name (str): Name of the mlp output torch module where attach the hook
        attn_out_proj_weight (str): Name of the attention output projection weight
        attn_out_proj_bias (str): Name of the attention output projection bias
        embed_tokens (str): Name of the embedding tokens torch module where attach the hook
        num_hidden_layers (int): Number of hidden layers
        num_attention_heads (int): Number of attention heads
        hidden_size (int): Hidden size of the transformer model
        num_key_value_heads (int): Number of key value heads
        num_key_value_groups (int): Number of key value groups
        head_dim (int): Dimension of the attention head

    """

    residual_stream_input_hook_name: str
    residual_stream_hook_name: str
    intermediate_stream_hook_name: str
    residual_stream_input_post_layernorm_hook_name: str
    head_key_hook_name: str
    head_value_hook_name: str
    head_query_hook_name: str
    # head_out_hook_name: str
    attn_in_hook_name: str
    attn_in_hook_name: str
    attn_out_hook_name: str
    attn_o_proj_input_hook_name: str
    attn_matrix_hook_name: str
    attn_matrix_pre_softmax_hook_name: str
    mlp_out_hook_name: str
    last_layernorm_hook_name: str


    attn_out_proj_weight: str
    attn_out_proj_bias: str
    embed_tokens: str
    unembed_matrix: str
    last_layernorm: str

    num_hidden_layers: int
    num_attention_heads: int
    hidden_size: int
    num_key_value_heads: int
    num_key_value_groups: int
    head_dim: int
    layernorm_type: Literal["RMS", "LayerNorm"]

    def use_language_model(self):
        """
        for the hook, remove the "language_model" prefix: "language_model.model.layers[{}]" -> "model.layers[{}]" . Remove from the data ckass
        """
        # iterate over the dataclass fields
        for field in self.__dataclass_fields__.keys():
            # remove the "language_model." prefix
            if "hook" in field:
                setattr(self, field, getattr(self, field).replace("language_model.", ""))

    def restore_full_model(self):
        for field in self.__dataclass_fields__.keys():
            if "hook" in field:
                setattr(self, field, getattr(self, field).replace("model.","language_model.model."))

use_language_model()

for the hook, remove the "language_model" prefix: "language_model.model.layers[{}]" -> "model.layers[{}]" . Remove from the data ckass

Source code in easyroutine/interpretability/models.py 
85
86
87
88
89
90
91
92
93
def use_language_model(self):
    """
    for the hook, remove the "language_model" prefix: "language_model.model.layers[{}]" -> "model.layers[{}]" . Remove from the data ckass
    """
    # iterate over the dataclass fields
    for field in self.__dataclass_fields__.keys():
        # remove the "language_model." prefix
        if "hook" in field:
            setattr(self, field, getattr(self, field).replace("language_model.", ""))

ModelFactory

This class is a factory to load the model and the processor. It supports the following models:

Supported Models

The following models are supported by this factory:

  • Chameleon-7b: A 7-billion parameter model for general-purpose tasks.
  • Chameleon-30b: A larger version of the Chameleon series with 30 billion parameters.
  • llava-hf/llava-v1.6-mistral-7b-hf: A 7-billion parameter model for multimodal tasks.
  • Pixtral-12b: Optimized for image-to-text tasks.
  • Emu3-Chat: Fine-tuned for conversational AI.
  • Emu3-Gen: Specialized in text generation tasks.
  • Emu3-Stage1: Pretrained for multi-stage training pipelines.
  • hf-internal-testing: A tiny model for internal testing purposes.
Adding a New Model

To add a new model: 1. Implement its logic in the load_model method. 2. Ensure it is correctly initialized and validated.

Source code in easyroutine/interpretability/models.py 
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
270
271
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
class ModelFactory:
    r"""
    This class is a factory to load the model and the processor. It supports the following models:

    Supported Models:
        The following models are supported by this factory:

        - **Chameleon-7b**: A 7-billion parameter model for general-purpose tasks.
        - **Chameleon-30b**: A larger version of the Chameleon series with 30 billion parameters.
        - **llava-hf/llava-v1.6-mistral-7b-hf**: A 7-billion parameter model for multimodal tasks.
        - **Pixtral-12b**: Optimized for image-to-text tasks.
        - **Emu3-Chat**: Fine-tuned for conversational AI.
        - **Emu3-Gen**: Specialized in text generation tasks.
        - **Emu3-Stage1**: Pretrained for multi-stage training pipelines.
        - **hf-internal-testing**: A tiny model for internal testing purposes.

    Adding a New Model:
        To add a new model:
        1. Implement its logic in the `load_model` method.
        2. Ensure it is correctly initialized and validated.
    """

    @staticmethod
    def load_model(
        model_name: str,
        attn_implementation: str,
        torch_dtype: torch.dtype,
        device_map: str,
    ) -> Tuple[torch.nn.Module, Optional[torch.nn.Module], ModelConfig]:
        r"""
        Load the model and its configuration based on the model name.

        Args:
            model_name (str): Name of the model to load.
            attn_implementation (str): Attention implementation type. (eager, flash-attn, sdp)
            torch_dtype (torch.dtype): Data type of the model.
            device_map (str): Device map for the model.

        Returns:
            model (HuggingFaceModel): Model instance.
            model_config (ModelConfig): Model configuration.
        """
        if attn_implementation != "eager":
            logger.warning(
                "ModelFactory: Using an attention type different from eager or custom eager could have unexpected behavior in some experiments!",
            )

        language_model = None
        if model_name in ["facebook/chameleon-7b", "facebook/chameleon-30b"]:
            model = ChameleonForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            model_config = ModelConfig(
                residual_stream_input_hook_name="model.layers[{}].input",
                residual_stream_hook_name="model.layers[{}].output",
                intermediate_stream_hook_name="model.layers[{}].post_attention_layernorm.output",
                residual_stream_input_post_layernorm_hook_name="model.layers[{}].self_attn.input",
                head_key_hook_name="model.layers[{}].self_attn.k_proj.output",
                head_value_hook_name="model.layers[{}].self_attn.v_proj.output",
                head_query_hook_name="model.layers[{}].self_attn.q_proj.output",
                attn_out_hook_name="model.layers[{}].self_attn.o_proj.output",
                attn_o_proj_input_hook_name="model.layers[{}].self_attn.o_proj.input",
                attn_in_hook_name="model.layers[{}].self_attn.input",
                mlp_out_hook_name="model.layers[{}].mlp.down_proj.output",
                attn_matrix_hook_name="model.layers[{}].self_attn.attention_matrix_hook.output",
                attn_matrix_pre_softmax_hook_name="model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
                last_layernorm_hook_name="model.norm.input",
                attn_out_proj_weight="model.layers[{}].self_attn.o_proj.weight",
                attn_out_proj_bias="model.layers[{}].self_attn.o_proj.bias",
                embed_tokens="model.layers[0].input",
                unembed_matrix="lm_head.weight",
                last_layernorm="model.norm",
                num_hidden_layers=model.config.num_hidden_layers,
                num_attention_heads=model.config.num_attention_heads,
                hidden_size=model.config.hidden_size,
                num_key_value_heads=model.config.num_key_value_heads,
                num_key_value_groups=model.config.num_attention_heads
                // model.config.num_key_value_heads,
                head_dim=model.config.hidden_size // model.config.num_attention_heads,
                layernorm_type="RMS",
            )

        elif model_name in [
            "mistral-community/pixtral-12b",
            "llava-hf/llava-v1.6-mistral-7b-hf",
        ]:
            if model_name == "mistral-community/pixtral-12b":
                model = LlavaForConditionalGeneration.from_pretrained(
                    model_name,
                    torch_dtype=torch_dtype,
                    device_map=device_map,
                    attn_implementation=attn_implementation,
                )
                model_config = ModelConfig(
                    residual_stream_input_hook_name="language_model.model.layers[{}].input",
                    residual_stream_hook_name="language_model.model.layers[{}].output",
                    intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
                    residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
                    head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
                    head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
                    head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
                    attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
                    attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
                    attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
                    attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
                    mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
                    last_layernorm_hook_name="language_model.model.norm.input",
                    attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
                    attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
                    embed_tokens="language_model.model.layers[0].input",
                    unembed_matrix="language_model.lm_head.weight",
                    last_layernorm="language_model.model.norm",
                    num_hidden_layers=model.language_model.config.num_hidden_layers,
                    num_attention_heads=model.language_model.config.num_attention_heads,
                    hidden_size=model.language_model.config.hidden_size,
                    num_key_value_heads=model.language_model.config.num_key_value_heads,
                    num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
                    head_dim=model.language_model.config.head_dim,
                    layernorm_type="RMS",
                )
            elif model_name == "llava-hf/llava-v1.6-mistral-7b-hf":
                model = LlavaNextForConditionalGeneration.from_pretrained(
                    model_name,
                    torch_dtype=torch_dtype,
                    device_map=device_map,
                    attn_implementation=attn_implementation,
                )
                model_config = ModelConfig(
                    residual_stream_input_hook_name="language_model.model.layers[{}].input",
                    residual_stream_hook_name="language_model.model.layers[{}].output",
                    intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
                    residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
                    head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
                    head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
                    head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
                    attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
                    attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
                    attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
                    attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
                    attn_matrix_pre_softmax_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
                    mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
                    last_layernorm_hook_name="language_model.model.norm.input",
                    attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
                    attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
                    embed_tokens="language_model.model.layers[0].input",
                    unembed_matrix="language_model.lm_head.weight",
                    last_layernorm="language_model.model.norm",
                    num_hidden_layers=model.language_model.config.num_hidden_layers,
                    num_attention_heads=model.language_model.config.num_attention_heads,
                    hidden_size=model.language_model.config.hidden_size,
                    num_key_value_heads=model.language_model.config.num_key_value_heads,
                    num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
                    head_dim=model.language_model.config.head_dim,
                    layernorm_type="RMS",
                )

            else:
                raise ValueError("Unsupported model_name")
            language_model = model.language_model

        elif model_name in ["Emu3-Chat", "Emu3-Gen", "Emu3-Stage1"]:
            raise NotImplementedError("Emu3 model not implemented yet")

        elif model_name in ["google/gemma-3-1b-it", "google/gemma-3-4b-it", "google/gemma-3-12b-it", "google/gemma-3-27b-it"]:
            model = Gemma3ForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            model_config = ModelConfig(
                residual_stream_input_hook_name="language_model.model.layers[{}].input",
                residual_stream_hook_name="language_model.model.layers[{}].output",
                intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
                residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
                head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
                head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
                head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
                attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
                attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
                attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
                attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
                attn_matrix_pre_softmax_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
                mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
                last_layernorm_hook_name="language_model.model.norm.input",
                attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
                attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
                embed_tokens="language_model.model.layers[0].input",
                unembed_matrix="language_model.lm_head.weight",
                last_layernorm="language_model.model.norm",
                num_hidden_layers=model.language_model.config.num_hidden_layers,
                num_attention_heads=model.language_model.config.num_attention_heads,
                hidden_size=model.language_model.config.hidden_size,
                num_key_value_heads=model.language_model.config.num_key_value_heads,
                num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
                head_dim=model.language_model.config.head_dim,
                layernorm_type="RMS",
            )
            language_model = model.language_model


        elif model_name in ["hf-internal-testing/tiny-random-LlamaForCausalLM"]:
            model = LlamaForCausalLM.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            model_config = ModelConfig(
                residual_stream_input_hook_name="model.layers[{}].input",
                residual_stream_hook_name="model.layers[{}].output",
                intermediate_stream_hook_name="model.layers[{}].post_attention_layernorm.output",
                residual_stream_input_post_layernorm_hook_name="model.layers[{}].self_attn.input",
                head_query_hook_name="model.layers[{}].self_attn.q_proj.output",
                head_value_hook_name="model.layers[{}].self_attn.v_proj.output",
                head_key_hook_name="model.layers[{}].self_attn.k_proj.output",
                attn_out_hook_name="model.layers[{}].self_attn.o_proj.output",
                attn_o_proj_input_hook_name="model.layers[{}].self_attn.o_proj.input",
                attn_in_hook_name="model.layers[{}].self_attn.input",
                attn_matrix_hook_name="model.layers[{}].self_attn.attention_matrix_hook.output",
                attn_matrix_pre_softmax_hook_name="model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
                mlp_out_hook_name="model.layers[{}].mlp.down_proj.output",
                last_layernorm_hook_name="model.norm.input",
                attn_out_proj_weight="model.layers[{}].self_attn.o_proj.weight",
                attn_out_proj_bias="model.layers[{}].self_attn.o_proj.bias",
                embed_tokens="model.layers[0].input",
                unembed_matrix="lm_head.weight",
                last_layernorm="model.norm",
                num_hidden_layers=model.config.num_hidden_layers,
                num_attention_heads=model.config.num_attention_heads,
                hidden_size=model.config.hidden_size,
                num_key_value_heads=model.config.num_key_value_heads,
                num_key_value_groups=model.config.num_attention_heads
                // model.config.num_key_value_heads,
                head_dim=model.config.hidden_size // model.config.num_attention_heads,
                layernorm_type="RMS",
            )

        elif model_name in ["CohereForAI/aya-101"]:
            model = T5ForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            language_model = None
            raise NotImplementedError("CohereForAI model not implemented yet")
            # model_config = ModelFactory._create_model_config(
            #     model.config, prefix="encoder."
            # )

        else:
            raise ValueError("Unsupported model_name")
        return model, language_model, model_config

    @staticmethod
    def _create_model_config(**kwargs):
        raise NotImplementedError("This method should be implemented in the if")

load_model(model_name, attn_implementation, torch_dtype, device_map) staticmethod

Load the model and its configuration based on the model name.

Parameters:

Name Type Description Default
model_name str

Name of the model to load.

 required
attn_implementation str

Attention implementation type. (eager, flash-attn, sdp)

 required
torch_dtype dtype

Data type of the model.

 required
device_map str

Device map for the model.

 required

Returns:

Name Type Description
model HuggingFaceModel

Model instance.
model_config ModelConfig

Model configuration.
Source code in easyroutine/interpretability/models.py 
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
270
271
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
@staticmethod
def load_model(
    model_name: str,
    attn_implementation: str,
    torch_dtype: torch.dtype,
    device_map: str,
) -> Tuple[torch.nn.Module, Optional[torch.nn.Module], ModelConfig]:
    r"""
    Load the model and its configuration based on the model name.

    Args:
        model_name (str): Name of the model to load.
        attn_implementation (str): Attention implementation type. (eager, flash-attn, sdp)
        torch_dtype (torch.dtype): Data type of the model.
        device_map (str): Device map for the model.

    Returns:
        model (HuggingFaceModel): Model instance.
        model_config (ModelConfig): Model configuration.
    """
    if attn_implementation != "eager":
        logger.warning(
            "ModelFactory: Using an attention type different from eager or custom eager could have unexpected behavior in some experiments!",
        )

    language_model = None
    if model_name in ["facebook/chameleon-7b", "facebook/chameleon-30b"]:
        model = ChameleonForConditionalGeneration.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
            attn_implementation=attn_implementation,
        )
        model_config = ModelConfig(
            residual_stream_input_hook_name="model.layers[{}].input",
            residual_stream_hook_name="model.layers[{}].output",
            intermediate_stream_hook_name="model.layers[{}].post_attention_layernorm.output",
            residual_stream_input_post_layernorm_hook_name="model.layers[{}].self_attn.input",
            head_key_hook_name="model.layers[{}].self_attn.k_proj.output",
            head_value_hook_name="model.layers[{}].self_attn.v_proj.output",
            head_query_hook_name="model.layers[{}].self_attn.q_proj.output",
            attn_out_hook_name="model.layers[{}].self_attn.o_proj.output",
            attn_o_proj_input_hook_name="model.layers[{}].self_attn.o_proj.input",
            attn_in_hook_name="model.layers[{}].self_attn.input",
            mlp_out_hook_name="model.layers[{}].mlp.down_proj.output",
            attn_matrix_hook_name="model.layers[{}].self_attn.attention_matrix_hook.output",
            attn_matrix_pre_softmax_hook_name="model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
            last_layernorm_hook_name="model.norm.input",
            attn_out_proj_weight="model.layers[{}].self_attn.o_proj.weight",
            attn_out_proj_bias="model.layers[{}].self_attn.o_proj.bias",
            embed_tokens="model.layers[0].input",
            unembed_matrix="lm_head.weight",
            last_layernorm="model.norm",
            num_hidden_layers=model.config.num_hidden_layers,
            num_attention_heads=model.config.num_attention_heads,
            hidden_size=model.config.hidden_size,
            num_key_value_heads=model.config.num_key_value_heads,
            num_key_value_groups=model.config.num_attention_heads
            // model.config.num_key_value_heads,
            head_dim=model.config.hidden_size // model.config.num_attention_heads,
            layernorm_type="RMS",
        )

    elif model_name in [
        "mistral-community/pixtral-12b",
        "llava-hf/llava-v1.6-mistral-7b-hf",
    ]:
        if model_name == "mistral-community/pixtral-12b":
            model = LlavaForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            model_config = ModelConfig(
                residual_stream_input_hook_name="language_model.model.layers[{}].input",
                residual_stream_hook_name="language_model.model.layers[{}].output",
                intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
                residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
                head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
                head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
                head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
                attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
                attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
                attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
                attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
                mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
                last_layernorm_hook_name="language_model.model.norm.input",
                attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
                attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
                embed_tokens="language_model.model.layers[0].input",
                unembed_matrix="language_model.lm_head.weight",
                last_layernorm="language_model.model.norm",
                num_hidden_layers=model.language_model.config.num_hidden_layers,
                num_attention_heads=model.language_model.config.num_attention_heads,
                hidden_size=model.language_model.config.hidden_size,
                num_key_value_heads=model.language_model.config.num_key_value_heads,
                num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
                head_dim=model.language_model.config.head_dim,
                layernorm_type="RMS",
            )
        elif model_name == "llava-hf/llava-v1.6-mistral-7b-hf":
            model = LlavaNextForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
                attn_implementation=attn_implementation,
            )
            model_config = ModelConfig(
                residual_stream_input_hook_name="language_model.model.layers[{}].input",
                residual_stream_hook_name="language_model.model.layers[{}].output",
                intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
                residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
                head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
                head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
                head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
                attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
                attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
                attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
                attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
                attn_matrix_pre_softmax_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
                mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
                last_layernorm_hook_name="language_model.model.norm.input",
                attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
                attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
                embed_tokens="language_model.model.layers[0].input",
                unembed_matrix="language_model.lm_head.weight",
                last_layernorm="language_model.model.norm",
                num_hidden_layers=model.language_model.config.num_hidden_layers,
                num_attention_heads=model.language_model.config.num_attention_heads,
                hidden_size=model.language_model.config.hidden_size,
                num_key_value_heads=model.language_model.config.num_key_value_heads,
                num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
                head_dim=model.language_model.config.head_dim,
                layernorm_type="RMS",
            )

        else:
            raise ValueError("Unsupported model_name")
        language_model = model.language_model

    elif model_name in ["Emu3-Chat", "Emu3-Gen", "Emu3-Stage1"]:
        raise NotImplementedError("Emu3 model not implemented yet")

    elif model_name in ["google/gemma-3-1b-it", "google/gemma-3-4b-it", "google/gemma-3-12b-it", "google/gemma-3-27b-it"]:
        model = Gemma3ForConditionalGeneration.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
            attn_implementation=attn_implementation,
        )
        model_config = ModelConfig(
            residual_stream_input_hook_name="language_model.model.layers[{}].input",
            residual_stream_hook_name="language_model.model.layers[{}].output",
            intermediate_stream_hook_name="language_model.model.layers[{}].post_attention_layernorm.output",
            residual_stream_input_post_layernorm_hook_name="language_model.model.layers[{}].self_attn.input",
            head_key_hook_name="language_model.model.layers[{}].self_attn.k_proj.output",
            head_value_hook_name="language_model.model.layers[{}].self_attn.v_proj.output",
            head_query_hook_name="language_model.model.layers[{}].self_attn.q_proj.output",
            attn_out_hook_name="language_model.model.layers[{}].self_attn.o_proj.output",
            attn_o_proj_input_hook_name="language_model.model.layers[{}].self_attn.o_proj.input",
            attn_in_hook_name="language_model.model.layers[{}].self_attn.input",
            attn_matrix_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_hook.output",
            attn_matrix_pre_softmax_hook_name="language_model.model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
            mlp_out_hook_name="language_model.model.layers[{}].mlp.down_proj.output",
            last_layernorm_hook_name="language_model.model.norm.input",
            attn_out_proj_weight="language_model.model.layers[{}].self_attn.o_proj.weight",
            attn_out_proj_bias="language_model.model.layers[{}].self_attn.o_proj.bias",
            embed_tokens="language_model.model.layers[0].input",
            unembed_matrix="language_model.lm_head.weight",
            last_layernorm="language_model.model.norm",
            num_hidden_layers=model.language_model.config.num_hidden_layers,
            num_attention_heads=model.language_model.config.num_attention_heads,
            hidden_size=model.language_model.config.hidden_size,
            num_key_value_heads=model.language_model.config.num_key_value_heads,
            num_key_value_groups=model.language_model.config.num_attention_heads // model.language_model.config.num_key_value_heads,
            head_dim=model.language_model.config.head_dim,
            layernorm_type="RMS",
        )
        language_model = model.language_model


    elif model_name in ["hf-internal-testing/tiny-random-LlamaForCausalLM"]:
        model = LlamaForCausalLM.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
            attn_implementation=attn_implementation,
        )
        model_config = ModelConfig(
            residual_stream_input_hook_name="model.layers[{}].input",
            residual_stream_hook_name="model.layers[{}].output",
            intermediate_stream_hook_name="model.layers[{}].post_attention_layernorm.output",
            residual_stream_input_post_layernorm_hook_name="model.layers[{}].self_attn.input",
            head_query_hook_name="model.layers[{}].self_attn.q_proj.output",
            head_value_hook_name="model.layers[{}].self_attn.v_proj.output",
            head_key_hook_name="model.layers[{}].self_attn.k_proj.output",
            attn_out_hook_name="model.layers[{}].self_attn.o_proj.output",
            attn_o_proj_input_hook_name="model.layers[{}].self_attn.o_proj.input",
            attn_in_hook_name="model.layers[{}].self_attn.input",
            attn_matrix_hook_name="model.layers[{}].self_attn.attention_matrix_hook.output",
            attn_matrix_pre_softmax_hook_name="model.layers[{}].self_attn.attention_matrix_pre_softmax_hook.output",
            mlp_out_hook_name="model.layers[{}].mlp.down_proj.output",
            last_layernorm_hook_name="model.norm.input",
            attn_out_proj_weight="model.layers[{}].self_attn.o_proj.weight",
            attn_out_proj_bias="model.layers[{}].self_attn.o_proj.bias",
            embed_tokens="model.layers[0].input",
            unembed_matrix="lm_head.weight",
            last_layernorm="model.norm",
            num_hidden_layers=model.config.num_hidden_layers,
            num_attention_heads=model.config.num_attention_heads,
            hidden_size=model.config.hidden_size,
            num_key_value_heads=model.config.num_key_value_heads,
            num_key_value_groups=model.config.num_attention_heads
            // model.config.num_key_value_heads,
            head_dim=model.config.hidden_size // model.config.num_attention_heads,
            layernorm_type="RMS",
        )

    elif model_name in ["CohereForAI/aya-101"]:
        model = T5ForConditionalGeneration.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
            attn_implementation=attn_implementation,
        )
        language_model = None
        raise NotImplementedError("CohereForAI model not implemented yet")
        # model_config = ModelFactory._create_model_config(
        #     model.config, prefix="encoder."
        # )

    else:
        raise ValueError("Unsupported model_name")
    return model, language_model, model_config

TokenizerFactory

This class return the right tokenizer for the model. If the model is multimodal return is_a_process == True

Source code in easyroutine/interpretability/models.py 
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
class TokenizerFactory:
    r"""
    This class return the right tokenizer for the model. If the model is multimodal return is_a_process == True
    """

    @staticmethod
    def load_tokenizer(model_name: str, torch_dtype: torch.dtype, device_map: str):
        r"""
        Load the tokenizer based on the model name.

        Args:
            model_name (str): Name of the model to load.
            torch_dtype (torch.dtype): Data type of the model.
            device_map (str): Device map for the model.

        Returns:
            processor (Tokenizer): Processor instance.
            is_a_processor (bool): True if the model is multimodal, False otherwise.
        """
        if model_name in ["facebook/chameleon-7b", "facebook/chameleon-30b"]:
            processor = ChameleonProcessor.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = True
        elif model_name in ["meta-llama/Llama-3.2-1B", "meta-llama/Llama-3.2-3B"]:
            processor = LlamaTokenizerFast.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = False
        elif model_name in ["google/gemma-3-1b-it", "google/gemma-3-4b-it", "google/gemma-3-12b-it", "google/gemma-3-27b-it"]:
            processor = Gemma3Processor.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = True
        elif model_name in ["mistral-community/pixtral-12b"]:
            processor = PixtralProcessor.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = True
        elif model_name in ["llava-hf/llava-v1.6-mistral-7b-hf"]:
            processor = LlavaNextProcessor.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = True
        elif model_name in ["Emu3-Chat", "Emu3-Gen", "Emu3-Stage1"]:
            raise NotImplementedError("Emu3 model not implemented yet")
        elif model_name in ["hf-internal-testing/tiny-random-LlamaForCausalLM"]:
            processor = LlamaTokenizer.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = False
        elif model_name in ["CohereForAI/aya-101"]:
            processor = T5TokenizerFast.from_pretrained(
                model_name,
                torch_dtype=torch_dtype,
                device_map=device_map,
            )
            is_a_processor = False

        else:
            raise ValueError("Unsupported model_name")

        return processor, is_a_processor

load_tokenizer(model_name, torch_dtype, device_map) staticmethod

Load the tokenizer based on the model name.

Parameters:

Name Type Description Default
model_name str

Name of the model to load.

 required
torch_dtype dtype

Data type of the model.

 required
device_map str

Device map for the model.

 required

Returns:

Name Type Description
processor Tokenizer

Processor instance.
is_a_processor bool

True if the model is multimodal, False otherwise.
Source code in easyroutine/interpretability/models.py 
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
@staticmethod
def load_tokenizer(model_name: str, torch_dtype: torch.dtype, device_map: str):
    r"""
    Load the tokenizer based on the model name.

    Args:
        model_name (str): Name of the model to load.
        torch_dtype (torch.dtype): Data type of the model.
        device_map (str): Device map for the model.

    Returns:
        processor (Tokenizer): Processor instance.
        is_a_processor (bool): True if the model is multimodal, False otherwise.
    """
    if model_name in ["facebook/chameleon-7b", "facebook/chameleon-30b"]:
        processor = ChameleonProcessor.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = True
    elif model_name in ["meta-llama/Llama-3.2-1B", "meta-llama/Llama-3.2-3B"]:
        processor = LlamaTokenizerFast.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = False
    elif model_name in ["google/gemma-3-1b-it", "google/gemma-3-4b-it", "google/gemma-3-12b-it", "google/gemma-3-27b-it"]:
        processor = Gemma3Processor.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = True
    elif model_name in ["mistral-community/pixtral-12b"]:
        processor = PixtralProcessor.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = True
    elif model_name in ["llava-hf/llava-v1.6-mistral-7b-hf"]:
        processor = LlavaNextProcessor.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = True
    elif model_name in ["Emu3-Chat", "Emu3-Gen", "Emu3-Stage1"]:
        raise NotImplementedError("Emu3 model not implemented yet")
    elif model_name in ["hf-internal-testing/tiny-random-LlamaForCausalLM"]:
        processor = LlamaTokenizer.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = False
    elif model_name in ["CohereForAI/aya-101"]:
        processor = T5TokenizerFast.from_pretrained(
            model_name,
            torch_dtype=torch_dtype,
            device_map=device_map,
        )
        is_a_processor = False

    else:
        raise ValueError("Unsupported model_name")

    return processor, is_a_processor