draft, wip, but that's the idea

src/transformers/models/reference_vlm/__init__.py

@@ -0,0 +1,27 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_reference_vlm import *
+    from .modeling_reference_vlm import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)

src/transformers/models/reference_vlm/modeling_reference_vlm.py

@@ -0,0 +1,376 @@
+
+from __future__ import annotations
+import copy
+from dataclasses import dataclass
+from typing import Optional, Union, Tuple, Callable
+
+import torch
+import torch.nn as nn
+
+from ...cache_utils import Cache, DynamicCache
+from ...configuration_utils import PretrainedConfig
+from ...generation import GenerationMixin
+from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask, create_masks_for_generate
+from ...modeling_flash_attention_utils import FlashAttentionKwargs
+from ...modeling_layers import GradientCheckpointingLayer
+from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutputWithPast
+from ...modeling_utils import PreTrainedModel, ALL_ATTENTION_FUNCTIONS
+from ...processing_utils import Unpack
+from ...utils import auto_docstring, can_return_tuple, is_torchdynamo_compiling, logging
+from ...utils.deprecation import deprecate_kwarg
+
+logger = logging.get_logger(__name__)
+
+# ===== Outputs =====
+@dataclass
+class ReferenceVLMOutputWithPast(BaseModelOutputWithPast):
+    image_hidden_states: Optional[torch.Tensor] = None
+    audio_hidden_states: Optional[torch.Tensor] = None
+
+@dataclass
+class ReferenceVLMCausalLMOutputWithPast(CausalLMOutputWithPast):
+    image_hidden_states: Optional[torch.Tensor] = None
+    audio_hidden_states: Optional[torch.Tensor] = None
+
+# ===== Small utils =====
+class RMSNorm(nn.Module):
+    def __init__(self, dim: int, eps: float = 1e-6):
+        super().__init__()
+        self.eps = eps
+        self.weight = nn.Parameter(torch.zeros(dim))
+    def forward(self, x):
+        out = x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
+        out = out * (1.0 + self.weight.float())
+        return out.type_as(x)
+
+def rotate_half(x):
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+
+# ===== Attention Placeholder =====
+
+def eager_attention_forward(module, q, k, v, attention_mask, dropout=0.0, scaling=None, sliding_window=None, **kwargs):
+    if scaling is None:
+        scaling = module.head_dim ** -0.5
+    k = repeat_kv(k, module.num_key_value_groups)
+    v = repeat_kv(v, module.num_key_value_groups)
+    attn_weights = torch.matmul(q, k.transpose(2,3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : k.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+    attn_weights = torch.softmax(attn_weights, dim=-1, dtype=torch.float32).to(q.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, v).transpose(1,2).contiguous()
+    return attn_output, attn_weights
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    b, n_kv, s, d = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(b, n_kv, n_rep, s, d)
+    return hidden_states.reshape(b, n_kv * n_rep, s, d)
+
+# ===== Image scatter helper (pure functional) =====
+
+def scatter_image_embeddings(inputs_embeds, image_features, special_image_mask):
+    image_features = image_features.to(inputs_embeds.device, inputs_embeds.dtype)
+    return inputs_embeds.masked_scatter(special_image_mask, image_features)
+
+# ===== Token-type OR-mask (Gemma3 style) =====
+
+def token_type_ids_mask_function(token_type_ids: Optional[torch.Tensor], image_group_ids: Optional[torch.Tensor], tokens_per_image: int) -> Optional[Callable]:
+    if token_type_ids is None:
+        return None
+    def inner_mask(batch_idx: int, head_idx: int, q_idx: int, kv_idx: int) -> bool:
+        safe_idx = torch.where(kv_idx < token_type_ids.shape[1], kv_idx, 0)
+        kv_tt = torch.where(kv_idx < token_type_ids.shape[1], token_type_ids[batch_idx, safe_idx], 0)
+        kv_gid = torch.where(kv_idx < image_group_ids.shape[1], image_group_ids[batch_idx, safe_idx], -1)
+        is_image_block = (token_type_ids[batch_idx, q_idx] == 1) & (kv_tt == 1)
+        same_image_block = image_group_ids[batch_idx, q_idx] == kv_gid
+        return is_image_block & same_image_block
+    return inner_mask
+
+# ===== Core Modules =====
+class ReferenceAttention(nn.Module):
+    def __init__(self, config, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = config.head_dim
+        self.num_heads = config.num_attention_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.scaling = config.query_pre_attn_scalar ** -0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_sliding = getattr(config.layer_types, '__getitem__', lambda i: 'full_attention')(layer_idx) == 'sliding_attention'
+        self.sliding_window = config.sliding_window if self.is_sliding else None
+        self.q_proj = nn.Linear(config.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias)
+        self.k_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        self.v_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, config.hidden_size, bias=config.attention_bias)
+        self.q_norm = RMSNorm(config.head_dim, eps=config.rms_norm_eps)
+        self.k_norm = RMSNorm(config.head_dim, eps=config.rms_norm_eps)
+        self.attn_logit_softcapping = getattr(config, 'attn_logit_softcapping', None)
+
+    def forward(self, hidden_states, position_embeddings, attention_mask=None, past_key_value: Optional[Cache]=None, cache_position=None, **kwargs: Unpack[FlashAttentionKwargs]):
+        bsz, q_len, _ = hidden_states.shape
+        q = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1,2)
+        k = self.k_proj(hidden_states).view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1,2)
+        v = self.v_proj(hidden_states).view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1,2)
+        q = self.q_norm(q); k = self.k_norm(k)
+        cos, sin = position_embeddings
+        q, k = apply_rotary_pos_emb(q, k, cos, sin)
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            k, v = past_key_value.update(k, v, self.layer_idx, cache_kwargs)
+        attn_impl = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] if self.config._attn_implementation != "eager" else eager_attention_forward
+        attn_out, attn_weights = attn_impl(self, q, k, v, attention_mask, dropout=self.attention_dropout if self.training else 0.0, scaling=self.scaling, sliding_window=self.sliding_window, **kwargs)
+        attn_out = attn_out.reshape(bsz, q_len, -1).contiguous()
+        attn_out = self.o_proj(attn_out)
+        return attn_out, attn_weights
+
+class ReferenceDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config, layer_idx):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.attention_type = config.layer_types[layer_idx] if hasattr(config, 'layer_types') else 'full_attention'
+        self.self_attn = ReferenceAttention(config, layer_idx)
+        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.pre_ffn_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_ffn_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.mlp = nn.Sequential(
+            nn.Linear(config.hidden_size, config.intermediate_size, bias=False),
+            nn.GELU(),
+            nn.Linear(config.intermediate_size, config.hidden_size, bias=False),
+        )
+
+    @deprecate_kwarg("last_cache_position", version="4.53.0")
+    def forward(self, hidden_states, position_embeddings_global, position_embeddings_local, attention_mask=None, position_ids=None, past_key_value: Optional[Cache]=None, output_attentions=False, use_cache=False, cache_position=None, **kwargs):
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        pos_emb = position_embeddings_local if self.self_attn.is_sliding else position_embeddings_global
+        hidden_states, attn_w = self.self_attn(hidden_states, pos_emb, attention_mask, past_key_value, cache_position, **kwargs)
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = residual + hidden_states
+        residual = hidden_states
+        hidden_states = self.pre_ffn_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = self.post_ffn_layernorm(hidden_states)
+        hidden_states = residual + hidden_states
+        outputs = (hidden_states,)
+        if output_attentions:
+            outputs += (attn_w,)
+        return outputs
+
+# ===== TEXT BACKBONE WRAPPER =====
+class ReferenceTextModel(PreTrainedModel):
+    config_class = None
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["ReferenceDecoderLayer"]
+    def __init__(self, config):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=self.padding_idx)
+        self.layers = nn.ModuleList([ReferenceDecoderLayer(config, i) for i in range(config.num_hidden_layers)])
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        # rotary emb placeholders (global & local like Gemma3)
+        self.rotary_emb = nn.Identity()
+        self.rotary_emb_local = nn.Identity()
+        self.gradient_checkpointing = False
+        self.post_init()
+
+    @can_return_tuple
+    def forward(self, input_ids=None, attention_mask=None, position_ids=None, past_key_values: Optional[Cache]=None, inputs_embeds=None, use_cache=None, output_attentions=None, output_hidden_states=None, cache_position=None, **kwargs):
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("Specify exactly one of input_ids or inputs_embeds")
+        if self.gradient_checkpointing and self.training and use_cache:
+            use_cache = False
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        if use_cache and past_key_values is None and not self.training:
+            past_key_values = DynamicCache()
+        if cache_position is None:
+            past_seen = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(past_seen, past_seen + inputs_embeds.shape[1], device=inputs_embeds.device)
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+        # causal masks
+        if not isinstance(attention_mask, dict):
+            mask_kwargs = dict(config=self.config, input_embeds=inputs_embeds, attention_mask=attention_mask, cache_position=cache_position, past_key_values=past_key_values, position_ids=position_ids)
+            causal_mask_mapping = {
+                "full_attention": create_causal_mask(**mask_kwargs),
+                "sliding_attention": create_sliding_window_causal_mask(**mask_kwargs),
+            }
+        else:
+            causal_mask_mapping = attention_mask
+        hidden_states = inputs_embeds
+        position_embeddings_global = self.rotary_emb(hidden_states, position_ids) if callable(getattr(self.rotary_emb, '__call__', None)) else (None,None)
+        position_embeddings_local = self.rotary_emb_local(hidden_states, position_ids) if callable(getattr(self.rotary_emb_local, '__call__', None)) else (None,None)
+        all_h = () if output_hidden_states else None
+        all_a = () if output_attentions else None
+        for layer in self.layers[: self.config.num_hidden_layers]:
+            if output_hidden_states:
+                all_h = all_h + (hidden_states,)
+            layer_out = layer(hidden_states, position_embeddings_global, position_embeddings_local, attention_mask=causal_mask_mapping[layer.attention_type], position_ids=position_ids, past_key_value=past_key_values, output_attentions=output_attentions, use_cache=use_cache, cache_position=cache_position, **kwargs)
+            hidden_states = layer_out[0]
+            if output_attentions:
+                all_a = all_a + (layer_out[1],)
+        hidden_states = self.norm(hidden_states)
+        if output_hidden_states:
+            all_h = all_h + (hidden_states,)
+        return BaseModelOutputWithPast(last_hidden_state=hidden_states, past_key_values=past_key_values, hidden_states=all_h, attentions=all_a)
+
+# ===== Causal LM head (text only) =====
+class ReferenceForCausalLM(ReferenceTextModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    base_model_prefix = "language_model"
+    def __init__(self, config):
+        super().__init__(config)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+    @can_return_tuple
+    def forward(self, input_ids=None, attention_mask=None, position_ids=None, past_key_values: Optional[Cache]=None, inputs_embeds=None, labels=None, use_cache=None, output_attentions=None, output_hidden_states=None, cache_position=None, logits_to_keep: Union[int, torch.Tensor]=0, **kwargs):
+        outputs = super().forward(input_ids=input_ids, attention_mask=attention_mask, position_ids=position_ids, past_key_values=past_key_values, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, cache_position=cache_position, **kwargs)
+        hs = outputs.last_hidden_state
+        slice_idx = slice(-logits_to_keep, None) if isinstance(logits_to_keep,int) else logits_to_keep
+        logits = self.lm_head(hs[:, slice_idx, :])
+        loss = None
+        if labels is not None:
+            logits = logits.float()
+            shift_logits = logits[..., :-1, :]
+            shift_labels = labels[..., 1:]
+            if attention_mask is not None:
+                shift_attention_mask = attention_mask[:, -shift_logits.shape[1]:].to(logits.device)
+                shift_logits = shift_logits[shift_attention_mask != 0].contiguous()
+                shift_labels = shift_labels[shift_attention_mask != 0].contiguous()
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, self.config.vocab_size), shift_labels.view(-1).to(shift_logits.device))
+        return CausalLMOutputWithPast(loss=loss, logits=logits, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions)
+
+# ===== Multimodal (text+vision(+audio)) =====
+class ReferenceVLMModel(PreTrainedModel):
+    accepts_loss_kwargs = False
+    def __init__(self, config):
+        super().__init__(config)
+        # plug real backbones
+        self.vision_tower = getattr(config, 'vision_model', None)
+        self.language_model = getattr(config, 'text_model', None)
+        self.multi_modal_projector = MultiModalProjector(config.vision_config.hidden_size, config.text_config.hidden_size)
+        self.vocab_size = config.text_config.vocab_size
+        self.post_init()
+    def get_input_embeddings(self):
+        return self.language_model.get_input_embeddings()
+    def set_input_embeddings(self, value):
+        self.language_model.set_input_embeddings(value)
+    def get_image_features(self, pixel_values: torch.Tensor):
+        vision_outputs = self.vision_tower(pixel_values=pixel_values).last_hidden_state
+        return self.multi_modal_projector(vision_outputs)
+    @can_return_tuple
+    def forward(self, input_ids=None, pixel_values=None, attention_mask=None, position_ids=None, past_key_values: Optional[Cache]=None, token_type_ids=None, cache_position=None, inputs_embeds=None, labels=None, use_cache=None, output_attentions=None, output_hidden_states=None, return_dict=None, **lm_kwargs):
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("Specify exactly one of input_ids or inputs_embeds")
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if input_ids is not None and self.config.image_token_id >= self.vocab_size:
+            special_image_mask = input_ids == self.config.image_token_id
+            llm_input_ids = input_ids.clone(); llm_input_ids[special_image_mask] = 0
+        else:
+            llm_input_ids = input_ids
+        if inputs_embeds is None:
+            inputs_embeds = self.get_input_embeddings()(llm_input_ids)
+        if cache_position is None:
+            past_seen = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(past_seen, past_seen + inputs_embeds.shape[1], device=inputs_embeds.device)
+        image_features = None
+        special_image_mask = None
+        if pixel_values is not None:
+            image_features = self.get_image_features(pixel_values)
+            if input_ids is None:
+                special_image_mask = inputs_embeds == self.get_input_embeddings()(torch.tensor(self.config.image_token_id, dtype=torch.long, device=inputs_embeds.device))
+                special_image_mask = special_image_mask.all(-1)
+            else:
+                special_image_mask = input_ids == self.config.image_token_id
+            special_image_mask = special_image_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
+            if not is_torchdynamo_compiling() and inputs_embeds[special_image_mask].numel() != image_features.numel():
+                raise ValueError("Mismatch between special image tokens and provided image features")
+            inputs_embeds = scatter_image_embeddings(inputs_embeds, image_features, special_image_mask)
+        if not isinstance(attention_mask, dict):
+            mask_kwargs = dict(config=self.config.get_text_config(), input_embeds=inputs_embeds, attention_mask=attention_mask, cache_position=cache_position, past_key_values=past_key_values, position_ids=position_ids)
+            if token_type_ids is not None and inputs_embeds.shape[1] != 1:
+                is_image = (token_type_ids == 1).to(cache_position.device)
+                new_image_start = is_image & ~nn.functional.pad(is_image, (1,0), value=0)[:, :-1]
+                image_group_ids = torch.cumsum(new_image_start.int(), dim=1) - 1
+                image_group_ids = torch.where(is_image, image_group_ids, torch.full_like(token_type_ids, -1))
+                mask_kwargs["or_mask_function"] = token_type_ids_mask_function(token_type_ids.to(cache_position.device), image_group_ids, self.config.mm_tokens_per_image)
+            causal_mask_mapping = {"full_attention": create_causal_mask(**mask_kwargs), "sliding_attention": create_sliding_window_causal_mask(**mask_kwargs)}
+        else:
+            causal_mask_mapping = attention_mask
+        outputs = self.language_model(attention_mask=causal_mask_mapping, position_ids=position_ids, past_key_values=past_key_values, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=True, cache_position=cache_position, **lm_kwargs)
+        return ReferenceVLMOutputWithPast(last_hidden_state=outputs.last_hidden_state, past_key_values=outputs.past_key_values if use_cache else None, hidden_states=outputs.hidden_states, attentions=outputs.attentions, image_hidden_states=image_features)
+
+class ReferenceVLMForConditionalGeneration(PreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = ReferenceVLMModel(config)
+        self.lm_head = nn.Linear(config.text_config.hidden_size, config.text_config.vocab_size, bias=False)
+        self.post_init()
+    def get_input_embeddings(self):
+        return self.model.get_input_embeddings()
+    def set_input_embeddings(self, value):
+        self.model.set_input_embeddings(value)
+    def forward(self, input_ids=None, pixel_values=None, attention_mask=None, position_ids=None, past_key_values: Optional[Cache]=None, token_type_ids=None, cache_position=None, inputs_embeds=None, labels=None, use_cache=None, output_attentions=None, output_hidden_states=None, return_dict=None, logits_to_keep: Union[int, torch.Tensor]=0, **lm_kwargs):
+        outputs = self.model(input_ids=input_ids, pixel_values=pixel_values, token_type_ids=token_type_ids, attention_mask=attention_mask, position_ids=position_ids, past_key_values=past_key_values, inputs_embeds=inputs_embeds, use_cache=use_cache, labels=labels, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, cache_position=cache_position, **lm_kwargs)
+        hs = outputs.last_hidden_state
+        slice_idx = slice(-logits_to_keep, None) if isinstance(logits_to_keep,int) else logits_to_keep
+        logits = self.lm_head(hs[:, slice_idx, :])
+        loss = None
+        if labels is not None:
+            logits = logits.float()
+            shift_logits = logits[..., :-1, :]
+            shift_labels = labels[..., 1:]
+            if attention_mask is not None:
+                mask2d = attention_mask[:, -shift_logits.shape[1]:].to(logits.device)
+                shift_logits = shift_logits[mask2d != 0].contiguous()
+                shift_labels = shift_labels[mask2d != 0].contiguous()
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, self.config.text_config.vocab_size), shift_labels.view(-1).to(shift_logits.device))
+        return ReferenceVLMCausalLMOutputWithPast(loss=loss, logits=logits, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions, image_hidden_states=outputs.image_hidden_states)
+    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, cache_position=None, position_ids=None, pixel_values=None, attention_mask=None, token_type_ids=None, use_cache=True, logits_to_keep=None, labels=None, **kwargs):
+        model_inputs = super().prepare_inputs_for_generation(input_ids, past_key_values=past_key_values, inputs_embeds=inputs_embeds, attention_mask=attention_mask, position_ids=position_ids, cache_position=cache_position, use_cache=use_cache, logits_to_keep=logits_to_keep, token_type_ids=token_type_ids, **kwargs)
+        if cache_position[0] == 0:
+            model_inputs["pixel_values"] = pixel_values
+        return model_inputs
+    @staticmethod
+    def create_masks_for_generate(config: PretrainedConfig, input_embeds: torch.Tensor, attention_mask: Optional[torch.Tensor], cache_position: torch.Tensor, past_key_values: Optional[Cache], position_ids: Optional[torch.Tensor], token_type_ids: Optional[torch.Tensor]=None, **kwargs) -> dict:
+        mask_kwargs = dict(config=config.get_text_config(), input_embeds=input_embeds, attention_mask=attention_mask, cache_position=cache_position, past_key_values=past_key_values, position_ids=position_ids)
+        if token_type_ids is not None and input_embeds.shape[1] != 1:
+            is_image = (token_type_ids == 1).to(cache_position.device)
+            new_image_start = is_image & ~nn.functional.pad(is_image, (1,0), value=0)[:, :-1]
+            image_group_ids = torch.cumsum(new_image_start.int(), dim=1) - 1
+            image_group_ids = torch.where(is_image, image_group_ids, torch.full_like(token_type_ids, -1))
+            mask_kwargs["or_mask_function"] = token_type_ids_mask_function(token_type_ids.to(cache_position.device), image_group_ids, config.mm_tokens_per_image)
+        return create_masks_for_generate(**mask_kwargs)
+
+__all__ = [
+    "ReferenceVLMOutputWithPast",
+    "ReferenceVLMCausalLMOutputWithPast",
+    "ReferenceAttention",
+    "ReferenceDecoderLayer",
+    "ReferenceTextModel",
+    "ReferenceForCausalLM",
+    "ReferenceVLMModel",
+    "ReferenceVLMForConditionalGeneration",
+]