You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1669 lines
76 KiB

# Copyright 2025 The ZhipuAI Inc. team and HuggingFace Inc. 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.
import itertools
from collections.abc import Callable
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import LayerNorm
from ... import initialization as init
from ...activations import ACT2FN
from ...cache_utils import Cache, DynamicCache
from ...configuration_utils import PreTrainedConfig
from ...feature_extraction_utils import BatchFeature
from ...image_utils import ImageInput
from ...masking_utils import create_causal_mask
from ...modeling_flash_attention_utils import FlashAttentionKwargs
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import BaseModelOutputWithPast, BaseModelOutputWithPooling
from ...modeling_rope_utils import RopeParameters
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from ...processing_utils import Unpack
from ...tokenization_utils_base import PreTokenizedInput, TextInput
from ...utils import (
TransformersKwargs,
auto_docstring,
can_return_tuple,
is_torchdynamo_compiling,
logging,
torch_compilable_check,
)
from ...utils.generic import check_model_inputs, maybe_autocast
from ...video_utils import VideoInput
from ..glm4.modeling_glm4 import Glm4MLP, Glm4RMSNorm, Glm4RotaryEmbedding, eager_attention_forward
from ..qwen2_5_vl.modeling_qwen2_5_vl import (
Qwen2_5_VisionPatchEmbed,
Qwen2_5_VisionRotaryEmbedding,
Qwen2_5_VLCausalLMOutputWithPast,
Qwen2_5_VLForConditionalGeneration,
Qwen2_5_VLMLP,
Qwen2_5_VLModel,
Qwen2_5_VLModelOutputWithPast,
Qwen2_5_VLPreTrainedModel,
Qwen2_5_VLTextModel,
Qwen2_5_VLVisionAttention,
Qwen2_5_VLVisionBlock,
)
from ..qwen2_vl.processing_qwen2_vl import (
Qwen2VLProcessor,
Qwen2VLProcessorKwargs,
)
logger = logging.get_logger(__name__)
class Glm4vVisionConfig(PreTrainedConfig):
r"""
This is the configuration class to store the configuration of a [`Glm4vVisionModel`]. It is used to instantiate an Glm4vVisionModel
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield
a similar configuration to that of
GLM-4.1V-9B-Thinking [THUDM/GLM-4.1V-9B-Thinking](https://huggingface.co/THUDM/GLM-4.1V-9B-Thinking).
Args:
depth (`int`, *optional*, defaults to 24):
Number of layers (depth) in the model.
hidden_size (`int`, *optional*, defaults to 1536):
Dimensionality of the encoder layers and the pooler layer.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
`"relu"`, `"selu"` and `"gelu_new"` are supported.
attention_bias (`bool`, *optional*, defaults to `False`):
Whether to add a bias to the queries, keys and values.
attention_dropout (`float`, *optional*, defaults to 0.0):
Dropout probability for attention weights.
num_heads (`<fill_type>`, *optional*, defaults to 12): <fill_docstring>
in_channels (`<fill_type>`, *optional*, defaults to 3): <fill_docstring>
image_size (`int` or `list[int]`, *optional*, defaults to 336):
The size (resolution) of each image.
patch_size (`int`, *optional*, defaults to 14):
The size (resolution) of each patch.
rms_norm_eps (`float`, *optional*, defaults to 1e-05):
The epsilon used by the rms normalization layers.
spatial_merge_size (`int`, *optional*, defaults to 2):
The size used for merging spatial dimensions.
temporal_patch_size (`int`, *optional*, defaults to 2):
The size used for patches along the temporal dimension.
out_hidden_size (`int`, *optional*, defaults to 4096):
The output hidden size of the vision model.
intermediate_size (`int`, *optional*, defaults to 13696):
Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
Example:
```python
>>> from transformers import Glm4vVisionConfig, Glm4vVisionModel
>>> # Initializing a Glm4vVisionConfig GLM-4.1V-9B style configuration
>>> configuration = Glm4vVisionConfig()
>>> # Initializing a model (with random weights) from the GLM-4.1V-9B configuration
>>> model = Glm4vVisionModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```"""
model_type = "glm4v_vision"
base_config_key = "vision_config"
def __init__(
self,
depth=24,
hidden_size=1536,
hidden_act="silu",
attention_bias=False,
attention_dropout=0.0,
num_heads=12,
in_channels=3,
image_size=336,
patch_size=14,
rms_norm_eps=1e-05,
spatial_merge_size=2,
temporal_patch_size=2,
out_hidden_size=4096,
intermediate_size=13696,
initializer_range=0.02,
**kwargs,
):
super().__init__(**kwargs)
self.depth = depth
self.hidden_size = hidden_size
self.hidden_act = hidden_act
self.num_heads = num_heads
self.in_channels = in_channels
self.image_size = image_size
self.patch_size = patch_size
self.spatial_merge_size = spatial_merge_size
self.temporal_patch_size = temporal_patch_size
self.out_hidden_size = out_hidden_size
self.intermediate_size = intermediate_size
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.attention_bias = attention_bias
self.attention_dropout = attention_dropout
class Glm4vTextConfig(PreTrainedConfig):
r"""
This is the configuration class to store the configuration of a [`Glm4vModel`]. It is used to instantiate a
GLM-4.1V model according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of
GLM-4.1V-9B-Thinking [THUDM/GLM-4.1V-9B-Thinking](https://huggingface.co/THUDM/GLM-4.1V-9B-Thinking).
Configuration objects inherit from [`PreTrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PreTrainedConfig`] for more information.
Args:
vocab_size (`int`, *optional*, defaults to 151552):
Vocabulary size of the Glm4v model. Defines the number of different tokens that can be represented by the
`inputs_ids` passed when calling [`Glm4vModel`]
hidden_size (`int`, *optional*, defaults to 4096):
Dimension of the hidden representations.
intermediate_size (`int`, *optional*, defaults to 13696):
Dimension of the MLP representations.
num_hidden_layers (`int`, *optional*, defaults to 40):
Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 32):
Number of attention heads for each attention layer in the Transformer encoder.
num_key_value_heads (`int`, *optional*, defaults to 2):
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
`num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
`num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
by meanpooling all the original heads within that group. For more details checkout [this
paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) in the decoder.
max_position_embeddings (`int`, *optional*, defaults to 32768):
The maximum sequence length that this model might ever be used with.
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
rms_norm_eps (`float`, *optional*, defaults to 1e-05):
The epsilon used by the rms normalization layers.
use_cache (`bool`, *optional*, defaults to `True`):
Whether or not the model should return the last key/values attentions (not used by all models). Only
relevant if `config.is_decoder=True`.
attention_dropout (`float`, *optional*, defaults to 0.0):
The dropout ratio for the attention probabilities.
rope_parameters (`RopeParameters`, *optional*):
Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain
a value for `rope_theta` and optionally parameters used for scaling in case you want to use RoPE
with longer `max_position_embeddings`.
pad_token_id (`int`, *optional*):
The id of the padding token.
```python
>>> from transformers import Glm4vTextModel, Glm4vConfig
>>> # Initializing a GLM-4.1V style configuration
>>> configuration = Glm4vConfig()
>>> # Initializing a model from the GLM-4.1V style configuration
>>> model = Glm4vTextModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```"""
model_type = "glm4v_text"
base_config_key = "text_config"
keys_to_ignore_at_inference = ["past_key_values"]
# Default tensor parallel plan for base model `Glm4v`
base_model_tp_plan = {
"layers.*.self_attn.q_proj": "colwise",
"layers.*.self_attn.k_proj": "colwise",
"layers.*.self_attn.v_proj": "colwise",
"layers.*.self_attn.o_proj": "rowwise",
"layers.*.mlp.gate_up_proj": "colwise_gather_output", # we need to replicate here due to the `chunk` operation
"layers.*.mlp.down_proj": "rowwise_split_input", # input is replicated due to the `chunk` operation
}
base_model_pp_plan = {
"embed_tokens": (["input_ids"], ["inputs_embeds"]),
"layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
"norm": (["hidden_states"], ["hidden_states"]),
}
def __init__(
self,
vocab_size: int | None = 151552,
hidden_size: int | None = 4096,
intermediate_size: int | None = 13696,
num_hidden_layers: int | None = 40,
num_attention_heads: int | None = 32,
num_key_value_heads: int | None = 2,
hidden_act: str | None = "silu",
max_position_embeddings: int | None = 32768,
initializer_range: float | None = 0.02,
rms_norm_eps: int | None = 1e-05,
use_cache: bool | None = True,
attention_dropout: float | None = 0.0,
rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
pad_token_id: int | None = None,
**kwargs,
):
self.vocab_size = vocab_size
self.max_position_embeddings = max_position_embeddings
self.hidden_size = hidden_size
self.intermediate_size = intermediate_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
# for backward compatibility
if num_key_value_heads is None:
num_key_value_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.hidden_act = hidden_act
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.use_cache = use_cache
self.attention_dropout = attention_dropout
self.rope_parameters = rope_parameters
self.pad_token_id = pad_token_id
super().__init__(ignore_keys_at_rope_validation={"mrope_section"}, **kwargs)
class Glm4vConfig(PreTrainedConfig):
r"""
This is the configuration class to store the configuration of a [`Glm4vModel`]. It is used to instantiate a
GLM-4.1V model according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of
GLM-4.1V-9B-Thinking [THUDM/GLM-4.1V-9B-Thinking](https://huggingface.co/THUDM/GLM-4.1V-9B-Thinking).
Configuration objects inherit from [`PreTrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PreTrainedConfig`] for more information.
Args:
text_config (`Union[PreTrainedConfig, dict]`, *optional*, defaults to `Glm4vTextConfig`):
The config object or dictionary of the text backbone.
vision_config (`Union[PreTrainedConfig, dict]`, *optional*, defaults to `Glm4vVisionConfig`):
The config object or dictionary of the vision backbone.
image_token_id (`int`, *optional*, defaults to 151343):
The image token index to encode the image prompt.
video_token_id (`int`, *optional*, defaults to 151344):
The video token index to encode the image prompt.
image_start_token_id (`int`, *optional*, defaults to 151339):
The image start token index to encode the start of image.
image_end_token_id (`int`, *optional*, defaults to 151340):
The image end token index to encode the end of image.
video_start_token_id (`int`, *optional*, defaults to 151341):
The video start token index to encode the start of video.
video_end_token_id (`int`, *optional*, defaults to 151342):
The video end token index to encode the end of video.
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
Whether the model's input and output word embeddings should be tied.
```python
>>> from transformers import Glm4vForConditionalGeneration, Glm4vConfig
>>> # Initializing a GLM-4.1V style configuration
>>> configuration = Glm4vConfig()
>>> # Initializing a model from the GLM-4.1V style configuration
>>> model = Glm4vForConditionalGeneration(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```"""
model_type = "glm4v"
sub_configs = {"vision_config": Glm4vVisionConfig, "text_config": Glm4vTextConfig}
keys_to_ignore_at_inference = ["past_key_values"]
def __init__(
self,
text_config=None,
vision_config=None,
image_token_id=151343,
video_token_id=151344,
image_start_token_id=151339,
image_end_token_id=151340,
video_start_token_id=151341,
video_end_token_id=151342,
tie_word_embeddings=False,
**kwargs,
):
if isinstance(vision_config, dict):
self.vision_config = self.sub_configs["vision_config"](**vision_config)
elif vision_config is None:
self.vision_config = self.sub_configs["vision_config"]()
if isinstance(text_config, dict):
self.text_config = self.sub_configs["text_config"](**text_config)
elif text_config is None:
self.text_config = self.sub_configs["text_config"](**kwargs)
self.image_token_id = image_token_id
self.video_token_id = video_token_id
self.video_start_token_id = video_start_token_id
self.video_end_token_id = video_end_token_id
self.image_start_token_id = image_start_token_id
self.image_end_token_id = image_end_token_id
self.tie_word_embeddings = tie_word_embeddings
super().__init__(**kwargs)
# Will be used for both Text and Vision modalities
class Glm4vRMSNorm(Glm4RMSNorm):
pass
class Glm4VisionMlp(Qwen2_5_VLMLP):
def __init__(self, config, bias: bool = False):
super().__init__(config, bias)
self.intermediate_size = config.out_hidden_size
class Glm4vVisionPatchEmbed(Qwen2_5_VisionPatchEmbed):
def __init__(self, config: Glm4vVisionConfig) -> None:
nn.Module.__init__(self)
self.patch_size = config.patch_size
self.temporal_patch_size = config.temporal_patch_size
self.in_channels = config.in_channels
self.embed_dim = config.hidden_size
kernel_size = [self.temporal_patch_size, self.patch_size, self.patch_size]
self.proj = nn.Conv3d(self.in_channels, self.embed_dim, kernel_size=kernel_size, stride=kernel_size)
class Glm4vVisionRotaryEmbedding(Qwen2_5_VisionRotaryEmbedding):
pass
class Glm4vVisionPatchMerger(nn.Module):
def __init__(self, dim: int, context_dim: int, hidden_act: str, bias: bool = False) -> None:
super().__init__()
self.proj = nn.Linear(dim, dim, bias=bias)
self.post_projection_norm = LayerNorm(dim)
self.gate_proj = nn.Linear(dim, context_dim, bias=bias)
self.up_proj = nn.Linear(dim, context_dim, bias=bias)
self.down_proj = nn.Linear(context_dim, dim, bias=bias)
self.act1 = nn.GELU()
self.act_fn = ACT2FN[hidden_act]
def forward(self, hidden_state: torch.Tensor) -> torch.Tensor:
hidden_state = self.proj(hidden_state)
hidden_state = self.act1(self.post_projection_norm(hidden_state))
return self.down_proj(self.act_fn(self.gate_proj(hidden_state)) * self.up_proj(hidden_state))
class Glm4vVisionEmbeddings(nn.Module):
def __init__(self, config: Glm4vVisionConfig):
super().__init__()
self.config = config
self.embed_dim = config.hidden_size
self.image_size = config.image_size
self.patch_size = config.patch_size
self.num_patches = (self.image_size // self.patch_size) ** 2
self.num_positions = self.num_patches
self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
self.interpolated_method = "bicubic"
def forward(self, embeddings, lengths, image_shapes, h_coords, w_coords) -> torch.Tensor:
"""
Forward pass with integrated position encoding adaptation using 2D interpolation.
Args:
embeddings: Input embeddings tensor
lengths (torch.Tensor): Sequence lengths for each image in the batch.
image_shapes (torch.Tensor): Tensor of shape [batch_size, 3] representing the image shapes (t, h, w).
h_coords (torch.Tensor): Tensor of shape [total_seq] representing the h coordinate for each patch.
w_coords (torch.Tensor): Tensor of shape [total_seq] representing the w coordinate for each patch.
Returns:
torch.Tensor: Embeddings with adapted position encoding added.
"""
# Get position embedding parameters
pos_embed_weight = self.position_embedding.weight
hidden_size = pos_embed_weight.shape[1]
device = pos_embed_weight.device
# Convert inputs to tensors if needed
if isinstance(lengths, list):
lengths = torch.tensor(lengths, device=device, dtype=torch.long)
# Prepare 2D position embedding
orig_size_sq = pos_embed_weight.shape[0]
orig_size = int(orig_size_sq**0.5)
pos_embed_2d = (
pos_embed_weight.view(orig_size, orig_size, hidden_size)
.permute(2, 0, 1)
.unsqueeze(0)
.to(device=device, dtype=torch.float32)
)
# Calculate target dimensions for each patch
target_h = torch.cat([image_shapes[i, 1].repeat(lengths[i]) for i in range(len(lengths))]).to(
device=device, dtype=torch.float32
)
target_w = torch.cat([image_shapes[i, 2].repeat(lengths[i]) for i in range(len(lengths))]).to(
device=device, dtype=torch.float32
)
# Normalize coordinates to [-1, 1] range for grid_sample
norm_w = ((w_coords + 0.5) / target_w) * 2 - 1
norm_h = ((h_coords + 0.5) / target_h) * 2 - 1
# Create sampling grid
grid = torch.stack((norm_w, norm_h), dim=-1).unsqueeze(0).unsqueeze(2)
# Perform bicubic interpolation
interpolated_embed_fp32 = F.grid_sample(
pos_embed_2d, grid, mode=self.interpolated_method, align_corners=False, padding_mode="border"
)
# Reshape and convert back to original dtype
adapted_pos_embed_fp32 = interpolated_embed_fp32.squeeze(0).squeeze(-1).permute(1, 0)
adapted_pos_embed = adapted_pos_embed_fp32.to(pos_embed_weight.dtype).to(embeddings.device)
# Add adapted position encoding to embeddings
embeddings = embeddings + adapted_pos_embed
return embeddings
class Glm4vVisionAttention(Qwen2_5_VLVisionAttention):
def __init__(self, config: Glm4vVisionConfig) -> None:
super().__init__(config)
self.attention_dropout = config.attention_dropout
self.qkv = nn.Linear(config.hidden_size, config.hidden_size * 3, bias=config.attention_bias)
self.proj = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
class Glm4vVisionBlock(Qwen2_5_VLVisionBlock):
def __init__(self, config) -> None:
super().__init__(config)
self.norm1 = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.norm2 = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.attn = Glm4vVisionAttention(config)
self.mlp = Glm4VisionMlp(config, bias=False)
class Glm4vTextRotaryEmbedding(Glm4RotaryEmbedding):
def __init__(self, config: Glm4vTextConfig, device=None):
super().__init__()
self.mrope_section = config.rope_parameters.get("mrope_section", [8, 12, 12])
def forward(self, x, position_ids):
# In contrast to other models, GLM-V has different position ids for the grids
# So we expand the inv_freq to shape (3, ...)
inv_freq_expanded = self.inv_freq[None, None, :, None].float().expand(3, position_ids.shape[1], -1, 1)
position_ids_expanded = position_ids[:, :, None, :].float() # shape (3, bs, 1, positions)
device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
with maybe_autocast(device_type=device_type, enabled=False): # Force float32
freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(2, 3)
freqs = self.apply_mrope(freqs, self.mrope_section)
emb = torch.cat((freqs, freqs), dim=-1)
cos = emb.cos() * self.attention_scaling
sin = emb.sin() * self.attention_scaling
return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
def apply_mrope(self, freqs, mrope_section):
section = mrope_section
chunks = freqs.split(section, dim=-1)
result = torch.cat([chunk[i % 3] for i, chunk in enumerate(chunks)], dim=-1)
return result
def rotate_half_llm(x):
"""Rotates half the hidden dims of the input."""
x1 = x[..., 0::2]
x2 = x[..., 1::2]
return torch.stack((-x2, x1), dim=-1).flatten(-2)
def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
"""Applies Rotary Position Embedding to the query and key tensors.
Args:
q (`torch.Tensor`): The query tensor.
k (`torch.Tensor`): The key tensor.
cos (`torch.Tensor`): The cosine part of the rotary embedding.
sin (`torch.Tensor`): The sine part of the rotary embedding.
unsqueeze_dim (`int`, *optional*, defaults to 1):
The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
Returns:
`tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
"""
cos = cos.unsqueeze(unsqueeze_dim)
sin = sin.unsqueeze(unsqueeze_dim)
# Interleave them instead of usual shape
cos = cos[..., : cos.shape[-1] // 2].repeat_interleave(2, dim=-1)
sin = sin[..., : sin.shape[-1] // 2].repeat_interleave(2, dim=-1)
# Keep half or full tensor for later concatenation
rotary_dim = cos.shape[-1]
q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
# Apply rotary embeddings on the first half or full tensor
q_embed = (q_rot * cos) + (rotate_half_llm(q_rot) * sin)
k_embed = (k_rot * cos) + (rotate_half_llm(k_rot) * sin)
# Concatenate back to full shape
q_embed = torch.cat([q_embed, q_pass], dim=-1)
k_embed = torch.cat([k_embed, k_pass], dim=-1)
return q_embed, k_embed
class Glm4vTextAttention(nn.Module):
"""
Multi-headed attention from 'Attention Is All You Need' paper.
and "Generating Long Sequences with Sparse Transformers".
"""
def __init__(self, config: Glm4vTextConfig, layer_idx: int | None = None):
super().__init__()
self.config = config
self.layer_idx = layer_idx
self.hidden_size = config.hidden_size
self.num_heads = config.num_attention_heads
self.head_dim = self.hidden_size // self.num_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.is_causal = True
self.attention_dropout = config.attention_dropout
self.rope_parameters = config.rope_parameters
self.scaling = self.head_dim**-0.5
self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=True)
self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True)
self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True)
self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
attention_mask: torch.Tensor | None = None,
past_key_values: Cache | None = None,
cache_position: torch.LongTensor | None = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> tuple[torch.Tensor, torch.Tensor | None, tuple[torch.Tensor] | None]:
bsz, q_len, _ = hidden_states.size()
query_states = self.q_proj(hidden_states)
key_states = self.k_proj(hidden_states)
value_states = self.v_proj(hidden_states)
query_states = query_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
key_states = key_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
value_states = value_states.view(bsz, q_len, -1, self.head_dim).transpose(1, 2)
cos, sin = position_embeddings
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
if past_key_values is not None:
cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} # Specific to RoPE models
key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
self.config._attn_implementation, eager_attention_forward
)
attn_output, attn_weights = attention_interface(
self,
query_states,
key_states,
value_states,
attention_mask,
dropout=0.0 if not self.training else self.attention_dropout,
scaling=self.scaling,
**kwargs,
)
attn_output = attn_output.reshape(bsz, q_len, -1).contiguous()
attn_output = self.o_proj(attn_output)
return attn_output, attn_weights
class Glm4vTextMLP(Glm4MLP):
pass
class Glm4vTextDecoderLayer(GradientCheckpointingLayer):
def __init__(self, config: Glm4vTextConfig, layer_idx: int):
super().__init__()
self.hidden_size = config.hidden_size
self.self_attn = Glm4vTextAttention(config, layer_idx)
self.mlp = Glm4vTextMLP(config)
self.input_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_self_attn_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_mlp_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
@auto_docstring
def forward(
self,
hidden_states: torch.Tensor,
position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
attention_mask: torch.Tensor | None = None,
position_ids: torch.LongTensor | None = None,
past_key_values: Cache | None = None,
use_cache: bool | None = False,
cache_position: torch.LongTensor | None = None,
**kwargs,
) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
# Self Attention
hidden_states, _ = self.self_attn(
hidden_states=hidden_states,
position_embeddings=position_embeddings,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=past_key_values,
use_cache=use_cache,
cache_position=cache_position,
**kwargs,
)
hidden_states = self.post_self_attn_layernorm(hidden_states)
hidden_states = residual + hidden_states
# Fully Connected
residual = hidden_states
hidden_states = self.post_attention_layernorm(hidden_states)
hidden_states = self.mlp(hidden_states)
hidden_states = self.post_mlp_layernorm(hidden_states)
hidden_states = residual + hidden_states
return hidden_states
class Glm4vModelOutputWithPast(Qwen2_5_VLModelOutputWithPast):
pass
class Glm4vPreTrainedModel(Qwen2_5_VLPreTrainedModel):
_no_split_modules = ["Glm4vTextDecoderLayer", "Glm4vVisionBlock"]
_can_record_outputs = {
"hidden_states": Glm4vTextDecoderLayer,
"attentions": Glm4vTextAttention,
}
def _init_weights(self, module):
PreTrainedModel._init_weights(self, module)
if isinstance(module, Glm4vVisionRotaryEmbedding):
inv_freq = 1.0 / (module.theta ** (torch.arange(0, module.dim, 2, dtype=torch.float) / module.dim))
init.copy_(module.inv_freq, inv_freq)
class Glm4vVisionModel(Glm4vPreTrainedModel):
config: Glm4vVisionConfig
input_modalities = ("image", "video")
_no_split_modules = ["Glm4vVisionBlock"]
_can_record_outputs = {
"hidden_states": Glm4vVisionBlock,
"attentions": Glm4vVisionAttention,
}
def __init__(self, config) -> None:
super().__init__(config)
self.spatial_merge_size = config.spatial_merge_size
self.patch_size = config.patch_size
self.embeddings = Glm4vVisionEmbeddings(config)
self.patch_embed = Glm4vVisionPatchEmbed(config)
head_dim = config.hidden_size // config.num_heads
self.rotary_pos_emb = Glm4vVisionRotaryEmbedding(head_dim // 2)
self.blocks = nn.ModuleList([Glm4vVisionBlock(config) for _ in range(config.depth)])
self.merger = Glm4vVisionPatchMerger(
dim=config.out_hidden_size, context_dim=config.intermediate_size, hidden_act=config.hidden_act
)
self.post_conv_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.downsample = nn.Conv2d(
in_channels=config.hidden_size,
out_channels=config.out_hidden_size,
kernel_size=config.spatial_merge_size,
stride=config.spatial_merge_size,
)
self.post_layernorm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.gradient_checkpointing = False
self.post_init()
def rot_pos_emb(self, grid_thw):
pos_ids = []
for t, h, w in grid_thw:
hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w)
hpos_ids = hpos_ids.reshape(
h // self.spatial_merge_size,
self.spatial_merge_size,
w // self.spatial_merge_size,
self.spatial_merge_size,
)
hpos_ids = hpos_ids.permute(0, 2, 1, 3)
hpos_ids = hpos_ids.flatten()
wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1)
wpos_ids = wpos_ids.reshape(
h // self.spatial_merge_size,
self.spatial_merge_size,
w // self.spatial_merge_size,
self.spatial_merge_size,
)
wpos_ids = wpos_ids.permute(0, 2, 1, 3)
wpos_ids = wpos_ids.flatten()
pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1))
pos_ids = torch.cat(pos_ids, dim=0)
max_grid_size = grid_thw[:, 1:].max()
rotary_pos_emb_full = self.rotary_pos_emb(max_grid_size)
rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
return rotary_pos_emb, pos_ids
@check_model_inputs
@auto_docstring
def forward(
self, hidden_states: torch.Tensor, grid_thw: torch.Tensor, **kwargs: Unpack[TransformersKwargs]
) -> tuple | BaseModelOutputWithPooling:
r"""
hidden_states (`torch.Tensor` of shape `(seq_len, hidden_size)`):
The final hidden states of the model.
grid_thw (`torch.Tensor` of shape `(num_images_or_videos, 3)`):
The temporal, height and width of feature shape of each image in LLM.
Returns:
`torch.Tensor`: hidden_states.
"""
hidden_states = self.patch_embed(hidden_states)
hidden_states = self.post_conv_layernorm(hidden_states)
rotary_pos_emb, image_type_ids = self.rot_pos_emb(grid_thw)
emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
position_embeddings = (emb.cos(), emb.sin())
cu_seqlens = torch.repeat_interleave(grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]).cumsum(
dim=0,
# Select dtype based on the following factors:
# - FA2 requires that cu_seqlens_q must have dtype int32
# - torch.onnx.export requires that cu_seqlens_q must have same dtype as grid_thw
# See https://github.com/huggingface/transformers/pull/34852 for more information
dtype=grid_thw.dtype if torch.jit.is_tracing() else torch.int32,
)
cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
seqlens = (cu_seqlens[1:] - cu_seqlens[:-1]).tolist()
hidden_states = self.embeddings(
hidden_states,
seqlens,
grid_thw,
image_type_ids[:, 0].to(hidden_states.device),
image_type_ids[:, 1].to(hidden_states.device),
)
for blk in self.blocks:
hidden_states = blk(
hidden_states,
cu_seqlens=cu_seqlens,
position_embeddings=position_embeddings,
**kwargs,
)
hidden_states = self.post_layernorm(hidden_states)
hidden_states = hidden_states.view(
-1, self.spatial_merge_size, self.spatial_merge_size, hidden_states.shape[-1]
)
hidden_states = hidden_states.permute(0, 3, 1, 2)
hidden_states = self.downsample(hidden_states).view(-1, self.config.out_hidden_size)
merged_hidden_states = self.merger(hidden_states)
return BaseModelOutputWithPooling(
last_hidden_state=hidden_states,
pooler_output=merged_hidden_states,
)
class Glm4vTextModel(Qwen2_5_VLTextModel):
def __init__(self, config: Glm4vTextConfig):
super().__init__(config)
self.layers = nn.ModuleList(
[Glm4vTextDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
)
self.norm = Glm4vRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.rotary_emb = Glm4vTextRotaryEmbedding(config=config)
del self._attn_implementation
del self.has_sliding_layers
@auto_docstring
@check_model_inputs
def forward(
self,
input_ids: torch.LongTensor | None = None,
attention_mask: torch.Tensor | None = None,
position_ids: torch.LongTensor | None = None,
past_key_values: Cache | None = None,
inputs_embeds: torch.FloatTensor | None = None,
use_cache: bool | None = None,
cache_position: torch.LongTensor | None = None,
**kwargs: Unpack[FlashAttentionKwargs],
) -> tuple | BaseModelOutputWithPast:
if (input_ids is None) ^ (inputs_embeds is not None):
raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
# torch.jit.trace() doesn't support cache objects in the output
if use_cache and past_key_values is None and not torch.jit.is_tracing():
past_key_values = DynamicCache(config=self.config)
if inputs_embeds is None:
inputs_embeds = self.embed_tokens(input_ids)
if cache_position is None:
past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
cache_position = torch.arange(
past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
)
# the hard coded `3` is for temporal, height and width.
if position_ids is None:
position_ids = cache_position.view(1, 1, -1).expand(3, inputs_embeds.shape[0], -1)
elif position_ids.ndim == 2:
position_ids = position_ids[None, ...].expand(3, position_ids.shape[0], -1)
# NOTE: we need to pass text position ids for packing. Qwen2-VL uses 3D positions
# where each dim indicates visual spatial positions for temporal/height/width grids.
# There are two scenarios when FA2-like packed masking might be activated.
# 1. User specifically passed packed `position_ids` and no attention mask.
# In this case we expect the useer to create correct position ids for all 3 grids
# and prepend text-only position ids to it. The final tensor will be [4, bs, seq-len]
# 2. User runs forward with no attention mask and no position ids. In this case, position ids
# are prepared by the model (`get_rope_index`) as `[4, bs, seq-len]` tensor. Text-only positions are
# prepended by us when creating positions so that the mask is constructed correctly. NOTE: failing to pass
# text-only positions will cause incorrect mask construction, do not change `prepare_input_for_generation`
if position_ids.ndim == 3 and position_ids.shape[0] == 4:
text_position_ids = position_ids[0]
position_ids = position_ids[1:]
else:
# If inputs are not packed (usual 3D positions), do not prepare mask from position_ids
text_position_ids = None
mask_kwargs = {
"config": self.config,
"input_embeds": inputs_embeds,
"attention_mask": attention_mask,
"cache_position": cache_position,
"past_key_values": past_key_values,
"position_ids": text_position_ids,
}
# Create the masks
causal_mask = create_causal_mask(**mask_kwargs)
hidden_states = inputs_embeds
position_embeddings = self.rotary_emb(hidden_states, position_ids=position_ids)
for decoder_layer in self.layers:
layer_outputs = decoder_layer(
hidden_states,
attention_mask=causal_mask,
position_ids=text_position_ids,
past_key_values=past_key_values,
cache_position=cache_position,
position_embeddings=position_embeddings,
**kwargs,
)
hidden_states = layer_outputs
hidden_states = self.norm(hidden_states)
return BaseModelOutputWithPast(
last_hidden_state=hidden_states,
past_key_values=past_key_values,
)
class Glm4vModel(Qwen2_5_VLModel):
_checkpoint_conversion_mapping = {}
_no_split_modules = ["Glm4vTextDecoderLayer", "Glm4vVisionBlock"]
def __init__(self, config):
super().__init__(config)
self.visual = Glm4vVisionModel._from_config(config.vision_config)
def get_rope_index(
self,
input_ids: torch.LongTensor | None = None,
image_grid_thw: torch.LongTensor | None = None,
video_grid_thw: torch.LongTensor | None = None,
attention_mask: torch.Tensor | None = None,
) -> tuple[torch.Tensor, torch.Tensor]:
"""
Calculate the 3D rope index based on image and video's temporal, height and width in LLM.
Explanation:
Each embedding sequence contains vision embedding and text embedding or just contains text embedding.
For pure text embedding sequence, the rotary position embedding has no difference with modern LLMs.
Examples:
input_ids: [T T T T T], here T is for text.
temporal position_ids: [0, 1, 2, 3, 4]
height position_ids: [0, 1, 2, 3, 4]
width position_ids: [0, 1, 2, 3, 4]
For vision and text embedding sequence, we calculate 3D rotary position embedding for vision part
and 1D rotary position embedding for text part.
Examples:
Temporal (Time): 3 patches, representing different segments of the video in time.
Height: 2 patches, dividing each frame vertically.
Width: 2 patches, dividing each frame horizontally.
We also have some important parameters:
fps (Frames Per Second): The video's frame rate, set to 1. This means one frame is processed each second.
tokens_per_second: This is a crucial parameter. It dictates how many "time-steps" or "temporal tokens" are conceptually packed into a one-second interval of the video. In this case, we have 25 tokens per second. So each second of the video will be represented with 25 separate time points. It essentially defines the temporal granularity.
temporal_patch_size: The number of frames that compose one temporal patch. Here, it's 2 frames.
interval: The step size for the temporal position IDs, calculated as tokens_per_second * temporal_patch_size / fps. In this case, 25 * 2 / 1 = 50. This means that each temporal patch will be have a difference of 50 in the temporal position IDs.
input_ids: [V V V V V V V V V V V V T T T T T], here V is for vision.
vision temporal position_ids: [0, 0, 0, 0, 50, 50, 50, 50, 100, 100, 100, 100]
vision height position_ids: [0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1]
vision width position_ids: [0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1]
text temporal position_ids: [101, 102, 103, 104, 105]
text height position_ids: [101, 102, 103, 104, 105]
text width position_ids: [101, 102, 103, 104, 105]
Here we calculate the text start position_ids as the max vision position_ids plus 1.
Args:
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.
image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
The temporal, height and width of feature shape of each image in LLM.
video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
The temporal, height and width of feature shape of each video in LLM.
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
- 1 for tokens that are **not masked**,
- 0 for tokens that are **masked**.
Returns:
position_ids (`torch.LongTensor` of shape `(3, batch_size, sequence_length)`)
mrope_position_deltas (`torch.Tensor` of shape `(batch_size)`)
"""
spatial_merge_size = self.config.vision_config.spatial_merge_size
image_token_id = self.config.image_token_id
video_start_token_id = self.config.video_start_token_id
video_end_token_id = self.config.video_end_token_id
mrope_position_deltas = []
if input_ids is not None and (image_grid_thw is not None or video_grid_thw is not None):
total_input_ids = input_ids
if attention_mask is None:
attention_mask = torch.ones_like(total_input_ids)
position_ids = torch.ones(
3,
input_ids.shape[0],
input_ids.shape[1],
dtype=input_ids.dtype,
device=input_ids.device,
)
image_index, video_index = 0, 0
video_group_index = 0
attention_mask = attention_mask.to(total_input_ids.device)
for i, input_ids in enumerate(total_input_ids):
input_ids = input_ids[attention_mask[i] == 1]
input_tokens = input_ids.tolist()
input_token_type = []
video_check_flg = False
for token in input_tokens:
if token == video_start_token_id:
video_check_flg = True
elif token == video_end_token_id:
video_check_flg = False
if token == image_token_id and not video_check_flg:
input_token_type.append("image")
elif token == image_token_id and video_check_flg:
input_token_type.append("video")
else:
input_token_type.append("text")
input_type_group = []
for key, group in itertools.groupby(enumerate(input_token_type), lambda x: x[1]):
group = list(group)
start_index = group[0][0]
end_index = group[-1][0] + 1
input_type_group.append((key, start_index, end_index))
llm_pos_ids_list = []
video_frame_num = 1
for modality_type, start_idx, end_idx in input_type_group:
st_idx = llm_pos_ids_list[-1].max() + 1 if len(llm_pos_ids_list) > 0 else 0
if modality_type == "image":
t, h, w = (
image_grid_thw[image_index][0],
image_grid_thw[image_index][1],
image_grid_thw[image_index][2],
)
llm_grid_t, llm_grid_h, llm_grid_w = (
t.item(),
h.item() // spatial_merge_size,
w.item() // spatial_merge_size,
)
t_index = torch.arange(llm_grid_t).view(-1, 1).expand(-1, llm_grid_h * llm_grid_w).flatten()
h_index = torch.arange(llm_grid_h).view(1, -1, 1).expand(llm_grid_t, -1, llm_grid_w).flatten()
w_index = torch.arange(llm_grid_w).view(1, 1, -1).expand(llm_grid_t, llm_grid_h, -1).flatten()
llm_pos_ids_list.append(torch.stack([t_index, h_index, w_index]) + st_idx)
image_index += 1
video_frame_num = 1
elif modality_type == "video":
t, h, w = (
video_frame_num,
video_grid_thw[video_index][1],
video_grid_thw[video_index][2],
)
llm_grid_t, llm_grid_h, llm_grid_w = (
t,
h.item() // spatial_merge_size,
w.item() // spatial_merge_size,
)
for t_idx in range(llm_grid_t):
t_index = torch.tensor(t_idx).view(-1, 1).expand(-1, llm_grid_h * llm_grid_w).flatten()
h_index = torch.arange(llm_grid_h).view(1, -1, 1).expand(1, -1, llm_grid_w).flatten()
w_index = torch.arange(llm_grid_w).view(1, 1, -1).expand(1, llm_grid_h, -1).flatten()
llm_pos_ids_list.append(torch.stack([t_index, h_index, w_index]) + st_idx)
video_group_index += 1
if video_group_index >= video_grid_thw[video_index][0]:
video_index += 1
video_group_index = 0
video_frame_num += 1
else:
text_len = end_idx - start_idx
llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx)
video_frame_num = 1
llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
position_ids[..., i, attention_mask[i] == 1] = llm_positions.to(position_ids.device)
mrope_position_deltas.append(llm_positions.max() + 1 - len(total_input_ids[i]))
mrope_position_deltas = torch.tensor(mrope_position_deltas, device=input_ids.device).unsqueeze(1)
return position_ids, mrope_position_deltas
else:
if attention_mask is not None:
position_ids = attention_mask.long().cumsum(-1) - 1
position_ids.masked_fill_(attention_mask == 0, 1)
position_ids = position_ids.unsqueeze(0).expand(3, -1, -1).to(attention_mask.device)
max_position_ids = position_ids.max(0, keepdim=False)[0].max(-1, keepdim=True)[0]
mrope_position_deltas = max_position_ids + 1 - attention_mask.shape[-1]
else:
position_ids = (
torch.arange(input_ids.shape[1], device=input_ids.device)
.view(1, 1, -1)
.expand(3, input_ids.shape[0], -1)
)
mrope_position_deltas = torch.zeros(
[input_ids.shape[0], 1],
device=input_ids.device,
dtype=input_ids.dtype,
)
return position_ids, mrope_position_deltas
@can_return_tuple
@auto_docstring
def get_video_features(
self,
pixel_values_videos: torch.FloatTensor,
video_grid_thw: torch.LongTensor | None = None,
**kwargs: Unpack[TransformersKwargs],
) -> tuple | BaseModelOutputWithPooling:
r"""
pixel_values_videos (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)`):
The tensors corresponding to the input videos.
video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
The temporal, height and width of feature shape of each video in LLM.
"""
pixel_values_videos = pixel_values_videos.type(self.visual.dtype)
# reshape video_grid_thw -> [b, 3] -> [1, h, w] * frames
temp_frames_hw = []
for t, h, w in video_grid_thw:
repeated_row = torch.tensor([1, h.item(), w.item()]).unsqueeze(0).repeat(t, 1)
temp_frames_hw.append(repeated_row)
flattened_video_grid_thw = torch.cat(temp_frames_hw, dim=0)
vision_outputs = self.visual(
pixel_values_videos, grid_thw=flattened_video_grid_thw, return_dict=True, **kwargs
)
split_sizes = (video_grid_thw.prod(-1) // self.visual.spatial_merge_size**2).tolist()
video_embeds = torch.split(vision_outputs.pooler_output, split_sizes)
vision_outputs.pooler_output = video_embeds
return vision_outputs
def get_placeholder_mask(
self,
input_ids: torch.LongTensor,
inputs_embeds: torch.FloatTensor,
image_features: torch.FloatTensor | None = None,
video_features: torch.FloatTensor | None = None,
):
"""
Obtains multimodal placeholder mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
equal to the length of multimodal features. If the lengths are different, an error is raised.
"""
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)
special_video_mask = inputs_embeds == self.get_input_embeddings()(
torch.tensor(self.config.video_token_id, dtype=torch.long, device=inputs_embeds.device)
)
special_video_mask = special_video_mask.all(-1)
else:
# GLM-4.1V and GLM-4.5V special_video_mask is special_image_mask
special_image_mask = input_ids == self.config.image_token_id
special_video_mask = input_ids == self.config.image_token_id
n_image_tokens = special_image_mask.sum()
special_image_mask = special_image_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
if image_features is not None:
torch_compilable_check(
inputs_embeds[special_image_mask].numel() == image_features.numel(),
f"Image features and image tokens do not match, tokens: {n_image_tokens}, features: {image_features.shape[0]}",
)
n_video_tokens = special_video_mask.sum()
special_video_mask = special_video_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
if video_features is not None:
torch_compilable_check(
inputs_embeds[special_video_mask].numel() == video_features.numel(),
f"Video features and video tokens do not match, tokens: {n_video_tokens}, features: {video_features.shape[0]}",
)
return special_image_mask, special_video_mask
@auto_docstring
@can_return_tuple
def forward(
self,
input_ids: torch.LongTensor | None = None,
attention_mask: torch.Tensor | None = None,
position_ids: torch.LongTensor | None = None,
past_key_values: Cache | None = None,
inputs_embeds: torch.FloatTensor | None = None,
pixel_values: torch.Tensor | None = None,
pixel_values_videos: torch.FloatTensor | None = None,
image_grid_thw: torch.LongTensor | None = None,
video_grid_thw: torch.LongTensor | None = None,
rope_deltas: torch.LongTensor | None = None,
cache_position: torch.LongTensor | None = None,
**kwargs: Unpack[TransformersKwargs],
) -> tuple | Glm4vModelOutputWithPast:
r"""
image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
The temporal, height and width of feature shape of each image in LLM.
video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
The temporal, height and width of feature shape of each video in LLM.
rope_deltas (`torch.LongTensor` of shape `(batch_size, )`, *optional*):
The rope index difference between sequence length and multimodal rope.
"""
if (input_ids is None) ^ (inputs_embeds is not None):
raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
if inputs_embeds is None:
inputs_embeds = self.get_input_embeddings()(input_ids)
if pixel_values is not None:
image_embeds = self.get_image_features(pixel_values, image_grid_thw, return_dict=True).pooler_output
image_embeds = torch.cat(image_embeds, dim=0).to(inputs_embeds.device, inputs_embeds.dtype)
image_mask, _ = self.get_placeholder_mask(input_ids, inputs_embeds, image_features=image_embeds)
inputs_embeds = inputs_embeds.masked_scatter(image_mask, image_embeds)
if pixel_values_videos is not None:
video_embeds = self.get_video_features(pixel_values_videos, video_grid_thw, return_dict=True).pooler_output
video_embeds = torch.cat(video_embeds, dim=0).to(inputs_embeds.device, inputs_embeds.dtype)
_, video_mask = self.get_placeholder_mask(input_ids, inputs_embeds, video_features=video_embeds)
inputs_embeds = inputs_embeds.masked_scatter(video_mask, video_embeds)
if position_ids is None:
attention_mask_tensor = (
attention_mask if not isinstance(attention_mask, dict) else attention_mask["full_attention"]
)
if attention_mask_tensor is not None and attention_mask_tensor.ndim == 4:
attention_mask_tensor = torch.diagonal(attention_mask_tensor[:, 0], dim1=1, dim2=2)
# Only apply conversion for floating point tensors (inverted masks)
if attention_mask_tensor.dtype.is_floating_point:
attention_mask_tensor = attention_mask_tensor / torch.finfo(attention_mask_tensor.dtype).min
attention_mask_tensor = (1.0 - attention_mask_tensor).int()
# Calculate RoPE index once per generation in the pre-fill stage only.
# When compiling, we can't check tensor values thus we check only input length
# It is safe to assume that `length!=1` means we're in pre-fill because compiled
# models currently cannot do asssisted decoding
prefill_compiled_stage = is_torchdynamo_compiling() and (
(input_ids is not None and input_ids.shape[1] != 1)
or (inputs_embeds is not None and inputs_embeds.shape[1] != 1)
)
prefill_noncompiled_stage = not is_torchdynamo_compiling() and (
(cache_position is not None and cache_position[0] == 0)
or (past_key_values is None or past_key_values.get_seq_length() == 0)
)
if (prefill_compiled_stage or prefill_noncompiled_stage) or self.rope_deltas is None:
position_ids, rope_deltas = self.get_rope_index(
input_ids,
image_grid_thw,
video_grid_thw,
attention_mask=attention_mask_tensor,
)
self.rope_deltas = rope_deltas
# then use the prev pre-calculated rope-deltas to get the correct position ids
else:
batch_size, seq_length, _ = inputs_embeds.shape
delta = (
(cache_position[0] + self.rope_deltas).to(inputs_embeds.device)
if cache_position is not None
else 0
)
position_ids = torch.arange(seq_length, device=inputs_embeds.device)
position_ids = position_ids.view(1, -1).expand(batch_size, -1)
if cache_position is not None: # otherwise `deltas` is an int `0`
delta = delta.repeat_interleave(batch_size // delta.shape[0], dim=0)
position_ids = position_ids.add(delta)
position_ids = position_ids.unsqueeze(0).expand(3, -1, -1)
outputs = self.language_model(
input_ids=None,
position_ids=position_ids,
attention_mask=attention_mask,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
cache_position=cache_position,
**kwargs,
)
return Glm4vModelOutputWithPast(
last_hidden_state=outputs.last_hidden_state,
past_key_values=outputs.past_key_values,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
rope_deltas=self.rope_deltas,
)
class Glm4vCausalLMOutputWithPast(Qwen2_5_VLCausalLMOutputWithPast):
pass
class Glm4vForConditionalGeneration(Qwen2_5_VLForConditionalGeneration):
_checkpoint_conversion_mapping = {}
def forward(
self,
input_ids: torch.LongTensor | None = None,
attention_mask: torch.Tensor | None = None,
position_ids: torch.LongTensor | None = None,
past_key_values: Cache | None = None,
inputs_embeds: torch.FloatTensor | None = None,
labels: torch.LongTensor | None = None,
pixel_values: torch.Tensor | None = None,
pixel_values_videos: torch.FloatTensor | None = None,
image_grid_thw: torch.LongTensor | None = None,
video_grid_thw: torch.LongTensor | None = None,
cache_position: torch.LongTensor | None = None,
logits_to_keep: int | torch.Tensor = 0,
**kwargs: Unpack[TransformersKwargs],
) -> tuple | Glm4vCausalLMOutputWithPast:
r"""
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
(masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*):
The temporal, height and width of feature shape of each image in LLM.
video_grid_thw (`torch.LongTensor` of shape `(num_videos, 3)`, *optional*):
The temporal, height and width of feature shape of each video in LLM.
Example:
```python
>>> from PIL import Image
>>> import httpx
>>> from io import BytesIO
>>> from transformers import AutoProcessor, Glm4vForConditionalGeneration
>>> model = Glm4vForConditionalGeneration.from_pretrained("zai-org/GLM-4.1V-9B-Thinking")
>>> processor = AutoProcessor.from_pretrained("zai-org/GLM-4.1V-9B-Thinking")
>>> messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://www.ilankelman.org/stopsigns/australia.jpg"},
{"type": "text", "text": "What is shown in this image?"},
],
},
]
>>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
>>> with httpx.stream("GET", url) as response:
... image = Image.open(BytesIO(response.read()))
>>> text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
>>> inputs = processor(text=[text], images=[image], vision_infos=[vision_infos])
>>> # Generate
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
"The image shows a street scene with a red stop sign in the foreground. In the background, there is a large red gate with Chinese characters ..."
```"""
outputs = self.model(
input_ids=input_ids,
pixel_values=pixel_values,
pixel_values_videos=pixel_values_videos,
image_grid_thw=image_grid_thw,
video_grid_thw=video_grid_thw,
position_ids=position_ids,
attention_mask=attention_mask,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
cache_position=cache_position,
**kwargs,
)
hidden_states = outputs[0]
# Only compute necessary logits, and do not upcast them to float if we are not computing the loss
slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
logits = self.lm_head(hidden_states[:, slice_indices, :])
loss = None
if labels is not None:
loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.text_config.vocab_size)
return Glm4vCausalLMOutputWithPast(
loss=loss,
logits=logits,
past_key_values=outputs.past_key_values,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
rope_deltas=outputs.rope_deltas,
)
def prepare_inputs_for_generation(
self,
input_ids,
past_key_values=None,
attention_mask=None,
inputs_embeds=None,
cache_position=None,
position_ids=None,
use_cache=True,
pixel_values=None,
pixel_values_videos=None,
image_grid_thw=None,
video_grid_thw=None,
is_first_iteration=False,
**kwargs,
):
# Overwritten -- in specific circumstances we don't want to forward image inputs to the model
model_inputs = super().prepare_inputs_for_generation(
input_ids,
past_key_values=past_key_values,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
cache_position=cache_position,
position_ids=position_ids,
pixel_values=pixel_values,
pixel_values_videos=pixel_values_videos,
image_grid_thw=image_grid_thw,
video_grid_thw=video_grid_thw,
use_cache=use_cache,
is_first_iteration=is_first_iteration,
**kwargs,
)
# GLM-V position_ids are prepared with rope_deltas in forward
model_inputs["position_ids"] = None
if not is_first_iteration and use_cache:
model_inputs["pixel_values"] = None
model_inputs["pixel_values_videos"] = None
return model_inputs
def _get_image_nums_and_video_nums(
self,
input_ids: torch.LongTensor | None,
inputs_embeds: torch.Tensor | None = None,
) -> tuple[torch.Tensor, torch.Tensor]:
"""
Get the number of images and videos for each sample to calculate the separation length of the sample tensor.
These parameters are not passed through the processor to avoid unpredictable impacts from interface modifications.
Args:
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
Indices of input sequence tokens in the vocabulary.
Returns:
image_nums (`torch.LongTensor` of shape `(batch_size, num_images_sample)`)
video_nums (`torch.LongTensor` of shape `(batch_size, num_videos_sample)`)
"""
if inputs_embeds is not None:
is_image = (
inputs_embeds
== self.get_input_embeddings()(
torch.tensor(self.config.image_start_token_id, dtype=torch.long, device=inputs_embeds.device)
)
)[..., 0]
is_video_start = (
inputs_embeds
== self.get_input_embeddings()(
torch.tensor(self.config.video_start_token_id, dtype=torch.long, device=inputs_embeds.device)
)
)[..., 0]
is_video_end = (
inputs_embeds
== self.get_input_embeddings()(
torch.tensor(self.config.video_end_token_id, dtype=torch.long, device=inputs_embeds.device)
)
)[..., 0]
else:
is_image = input_ids == self.config.image_start_token_id
is_video_start = input_ids == self.config.video_start_token_id
is_video_end = input_ids == self.config.video_end_token_id
# Cumulative sum to track if we're inside a video span
# We'll assume well-formed video tags (i.e. matching starts and ends)
video_level = torch.cumsum(is_video_start.int() - is_video_end.int(), dim=1)
inside_video = video_level > 0 # shape (batch_size, seq_length)
# Mask out image tokens that are inside video spans
standalone_images = is_image & (~inside_video)
# Count per batch
image_counts = standalone_images.sum(dim=1)
video_counts = is_video_start.sum(dim=1)
return image_counts, video_counts
class Glm4vProcessorKwargs(Qwen2VLProcessorKwargs):
_defaults = {
"text_kwargs": {
"padding": False,
"return_token_type_ids": False,
"return_mm_token_type_ids": False,
},
"videos_kwargs": {"return_metadata": True},
}
class Glm4vProcessor(Qwen2VLProcessor):
def __init__(self, image_processor=None, tokenizer=None, video_processor=None, chat_template=None, **kwargs):
super().__init__(image_processor, tokenizer, video_processor, chat_template=chat_template)
self.image_token = "<|image|>" if not hasattr(tokenizer, "image_token") else tokenizer.image_token
self.video_token = "<|video|>" if not hasattr(tokenizer, "video_token") else tokenizer.video_token
def __call__(
self,
images: ImageInput | None = None,
text: TextInput | PreTokenizedInput | list[TextInput] | list[PreTokenizedInput] = None,
videos: VideoInput | None = None,
**kwargs: Unpack[Glm4vProcessorKwargs],
) -> BatchFeature:
r"""
Returns:
[`BatchFeature`]: A [`BatchFeature`] with the following fields:
- **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
- **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
`return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
`None`).
- **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
- **pixel_values_videos** -- Pixel values of videos to be fed to a model. Returned when `videos` is not `None`.
- **image_grid_thw** -- List of image 3D grid in LLM. Returned when `images` is not `None`.
- **video_grid_thw** -- List of video 3D grid in LLM. Returned when `videos` is not `None`.
"""
output_kwargs = self._merge_kwargs(
Glm4vProcessorKwargs,
tokenizer_init_kwargs=self.tokenizer.init_kwargs,
**kwargs,
)
if images is not None:
image_inputs = self.image_processor(images=images, **output_kwargs["images_kwargs"])
image_grid_thw = image_inputs["image_grid_thw"]
else:
image_inputs = {}
image_grid_thw = None
if videos is not None:
videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"])
# If user has not requested video metadata, pop it
if not kwargs.get("return_metadata"):
video_metadata = videos_inputs.pop("video_metadata")
else:
video_metadata = videos_inputs["video_metadata"]
video_grid_thw = videos_inputs["video_grid_thw"]
else:
videos_inputs = {}
video_grid_thw = None
if not isinstance(text, list):
text = [text]
text = text.copy() # below lines change text in-place
if image_grid_thw is not None:
merge_length = self.image_processor.merge_size**2
index = 0
for i in range(len(text)):
while self.image_token in text[i]:
num_image_tokens = image_grid_thw[index].prod() // merge_length
text[i] = text[i].replace(self.image_token, "<|placeholder|>" * num_image_tokens, 1)
index += 1
text[i] = text[i].replace("<|placeholder|>", self.image_token)
if video_grid_thw is not None:
merge_length = self.video_processor.merge_size**2
video_index = 0
for i in range(len(text)):
while self.video_token in text[i]:
num_frames = video_grid_thw[video_index][0]
video_structure = ""
metadata = video_metadata[video_index]
if metadata.fps is None:
logger.warning_once(
"SmolVLM requires frame timestamps to construct prompts, but the `fps` of the input video could not be inferred. "
"Probably `video_metadata` was missing from inputs and you passed pre-sampled frames. "
"Defaulting to `fps=24`. Please provide `video_metadata` for more accurate results."
)
metadata.fps = 24 if metadata.fps is None else metadata.fps
timestamps = metadata.timestamps[::2] # mrope
unique_timestamps = []
for idx in range(0, len(timestamps)):
unique_timestamps.append(timestamps[idx])
selected_timestamps = unique_timestamps[:num_frames]
while len(selected_timestamps) < num_frames:
selected_timestamps.append(selected_timestamps[-1] if selected_timestamps else 0)
for frame_idx in range(num_frames):
timestamp_sec = selected_timestamps[frame_idx]
frame_structure = self.replace_frame_token_id(timestamp_sec)
video_structure += frame_structure
text[i] = text[i].replace(self.video_token, video_structure, 1)
num_image_tokens = (
video_grid_thw[video_index].prod() // merge_length // video_grid_thw[video_index][0]
)
for frame_idx in range(num_frames):
if self.image_token in text[i]:
text[i] = text[i].replace(self.image_token, "<|placeholder|>" * num_image_tokens, 1)
video_index += 1
text[i] = text[i].replace("<|placeholder|>", self.image_token)
return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None)
return_mm_token_type_ids = output_kwargs["text_kwargs"].pop("return_mm_token_type_ids", False)
text_inputs = self.tokenizer(text, **output_kwargs["text_kwargs"])
self._check_special_mm_tokens(text, text_inputs, modalities=["image", "video"])
if return_mm_token_type_ids:
array_ids = np.array(text_inputs["input_ids"])
mm_token_type_ids = np.zeros_like(text_inputs["input_ids"])
mm_token_type_ids[array_ids == self.image_token_id] = 1
text_inputs["mm_token_type_ids"] = mm_token_type_ids.tolist()
return BatchFeature(data={**text_inputs, **image_inputs, **videos_inputs}, tensor_type=return_tensors)
def replace_frame_token_id(self, timestamp_sec):
return f"<|begin_of_image|>{self.image_token}<|end_of_image|>{int(timestamp_sec)}"
__all__ = [
"Glm4vConfig",
"Glm4vTextConfig",
"Glm4vVisionConfig",
"Glm4vForConditionalGeneration",
"Glm4vModel",
"Glm4vPreTrainedModel",
"Glm4vProcessor",
"Glm4vTextModel",
"Glm4vVisionModel",
]