Transformers documentation
MiniCPM-V 4.7
This model was published in HF papers on 2025-09-16 and contributed to Hugging Face Transformers on 2026-09-21.
MiniCPM-V 4.7
MiniCPM-V is a series of efficient multimodal large language models developed by OpenBMB. Like MiniCPM-V 4.6, the MiniCPM-V 4.7 architecture pairs a SigLIP vision encoder that has a window-attention merger with a Qwen3.5 language model backbone, and supports both 4x and 16x visual downsampling modes.
The main addition over 4.6 is canvas M-RoPE: instead of numbering visual tokens along a single 1-D sequence, the model lays every image out on a 2-D canvas and assigns each visual token a (temporal, height, width) position, so slices of the same image keep their spatial relationship and video frames keep their temporal order.
This model was contributed by OpenBMB. The original code can be found here.
Passing
use_image_idto a processor will number several images in one prompt so the text can refer to them individually. It applies to images only: a video is a single temporal sequence of frames rather than several addressable visuals, which is how the model was trained, so the setting is ignored for video inputs.
Usage example
Inference with Pipeline
from transformers import pipeline
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg",
},
{"type": "text", "text": "Describe this image."},
],
},
]
pipe = pipeline("image-text-to-text", model="openbmb/MiniCPM-V-4_7")
outputs = pipe(text=messages, max_new_tokens=50, return_full_text=False)
outputs[0]["generated_text"]Inference on a single image
from transformers import AutoProcessor, AutoModelForImageTextToText
model_checkpoint = "openbmb/MiniCPM-V-4_7"
processor = AutoProcessor.from_pretrained(model_checkpoint)
model = AutoModelForImageTextToText.from_pretrained(model_checkpoint, device_map="auto")
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
{"type": "text", "text": "Describe this image."},
],
}
]
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt",
).to(model.device, dtype=model.dtype)
output = model.generate(**inputs, max_new_tokens=100)
decoded_output = processor.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True)
print(decoded_output)Downsampling mode
MiniCPM-V 4.7 supports two visual downsampling modes:
- 16x (default): More aggressive downsampling, fewer visual tokens, faster inference.
- 4x: Less downsampling, more visual tokens, better for detail-rich tasks.
You can change the downsampling mode at runtime by passing downsample_mode via processor_kwargs and to model.generate:
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt",
processor_kwargs={"downsample_mode": "4x"},
).to(model.device, dtype=model.dtype)
output = model.generate(**inputs, max_new_tokens=100, downsample_mode="4x")Thinking mode
The model supports a thinking mode controlled by enable_thinking in the chat template. When enabled, the model generates internal reasoning before providing the final answer:
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt",
enable_thinking=True,
).to(model.device, dtype=model.dtype)
output = model.generate(**inputs, max_new_tokens=1024)To disable thinking (default for evaluation):
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt",
enable_thinking=False,
).to(model.device, dtype=model.dtype)Video inference
MiniCPM-V 4.7 supports video understanding.
messages = [
{
"role": "user",
"content": [
{"type": "video", "video": "path/to/video.mp4"},
{"type": "text", "text": "Describe what happens in this video."},
],
}
]
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt",
).to(model.device, dtype=model.dtype)
output = model.generate(**inputs, max_new_tokens=200)
decoded_output = processor.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True)
print(decoded_output)MiniCPMV4_7Config
class transformers.MiniCPMV4_7Config
< source >( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: typing.Union[str, ForwardRef('torch.dtype'), NoneType] = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: typing.Optional[typing.Literal['regression', 'single_label_classification', 'multi_label_classification']] = Nonetext_config: dict | transformers.configuration_utils.PreTrainedConfig | None = Nonevision_config: dict | transformers.configuration_utils.PreTrainedConfig | None = Noneinsert_layer_id: int = 6image_size: int = 448image_token_id: int | None = Nonevideo_token_id: int | None = Nonetie_word_embeddings: bool = Falsedownsample_mode: str = '16x'merge_kernel_size: tuple[int, int] | list[int] = (2, 2)merger_times: int = 1image_start_id: int | None = Noneimage_end_id: int | None = Noneslice_start_id: int | None = Noneslice_end_id: int | None = Nonenewline_id: int | None = None )
Parameters
- text_config (
Union[dict, ~configuration_utils.PreTrainedConfig], optional) — The config object or dictionary of the text backbone. - vision_config (
Union[dict, ~configuration_utils.PreTrainedConfig], optional) — The config object or dictionary of the vision backbone. - insert_layer_id (
int, optional, defaults to 6) — Vision encoder layer index after which the window-attention merger is applied. - image_size (
int, optional, defaults to448) — The size (resolution) of each image. - image_token_id (
int, optional) — The image token index used as a placeholder for input images. - video_token_id (
int, optional) — The video token index used as a placeholder for input videos. - tie_word_embeddings (
bool, optional, defaults toFalse) — Whether to tie weight embeddings according to model’stied_weights_keysmapping. - downsample_mode (
str, optional, defaults to"16x") — Visual token downsampling ratio."4x"keeps 4× more tokens. - merge_kernel_size (
tuple[int, int], optional, defaults to(2, 2)) — Kernel size(h, w)for merging adjacent visual patches in the Merger. - merger_times (
int, optional, defaults to 1) — Number of iterative merge rounds in the Merger. - image_start_id (
int, optional) — Token id of the image-start marker (<image>). Canvas M-RoPE pins it to the halo just outside the top-left corner of the image canvas. Resolved from the tokenizer by the conversion script and stored inconfig.json; required for image or video inputs. - image_end_id (
int, optional) — Token id of the image-end marker (</image>). Canvas M-RoPE pins it to the far corner of the image canvas. Resolved from the tokenizer by the conversion script and stored inconfig.json; required for image or video inputs. - slice_start_id (
int, optional) — Token id of the slice-start marker (<slice>). Canvas M-RoPE pins it to the top-left corner of the slice it opens. Resolved from the tokenizer by the conversion script and stored inconfig.json; required for image or video inputs. - slice_end_id (
int, optional) — Token id of the slice-end marker (</slice>). Canvas M-RoPE pins it to the bottom-right corner of the slice it closes. Resolved from the tokenizer by the conversion script and stored inconfig.json; required for image or video inputs. - newline_id (
int, optional) — Token id of the newline ("\n") that separates slice rows. Canvas M-RoPE pins it just past the right edge of the row it ends. Resolved from the tokenizer by the conversion script and stored inconfig.json; required for image or video inputs.
This is the configuration class to store the configuration of a MiniCPMV4_7Model. It is used to instantiate a Minicpmv4 7 model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the openbmb/MiniCPM-V-4.7
Configuration objects inherit from PreTrainedConfig and can be used to control the model outputs. Read the documentation from PreTrainedConfig for more information.
MiniCPMV4_7VisionConfig
class transformers.MiniCPMV4_7VisionConfig
< source >( transformers_version: str | None = Nonearchitectures: list[str] | None = Noneoutput_hidden_states: bool | None = Falsereturn_dict: bool | None = Truedtype: typing.Union[str, ForwardRef('torch.dtype'), NoneType] = Nonechunk_size_feed_forward: int = 0is_encoder_decoder: bool = Falseid2label: dict[int, str] | dict[str, str] | None = Nonelabel2id: dict[str, int] | dict[str, str] | None = Noneproblem_type: typing.Optional[typing.Literal['regression', 'single_label_classification', 'multi_label_classification']] = Nonehidden_size: int = 768intermediate_size: int = 3072num_hidden_layers: int = 12num_attention_heads: int = 12num_channels: int = 3image_size: int | list[int] | tuple[int, int] = 224patch_size: int | list[int] | tuple[int, int] = 16hidden_act: str = 'gelu_pytorch_tanh'layer_norm_eps: float = 1e-06attention_dropout: float | int = 0.0insert_layer_id: int = 6window_kernel_size: tuple[int, int] | list[int] = (2, 2) )
Parameters
- hidden_size (
int, optional, defaults to768) — Dimension of the hidden representations. - intermediate_size (
int, optional, defaults to3072) — Dimension of the MLP representations. - num_hidden_layers (
int, optional, defaults to12) — Number of hidden layers in the Transformer decoder. - num_attention_heads (
int, optional, defaults to12) — Number of attention heads for each attention layer in the Transformer decoder. - num_channels (
int, optional, defaults to3) — The number of input channels. - image_size (
Union[int, list[int], tuple[int, int]], optional, defaults to224) — The size (resolution) of each image. - patch_size (
Union[int, list[int], tuple[int, int]], optional, defaults to16) — The size (resolution) of each patch. - hidden_act (
str, optional, defaults togelu_pytorch_tanh) — The non-linear activation function (function or string) in the decoder. For example,"gelu","relu","silu", etc. - layer_norm_eps (
float, optional, defaults to1e-06) — The epsilon used by the layer normalization layers. - attention_dropout (
Union[float, int], optional, defaults to0.0) — The dropout ratio for the attention probabilities. - insert_layer_id (
int, optional, defaults to 6) — Vision encoder layer index after which the window-attention merger is applied. - window_kernel_size (
tuple[int, int], optional, defaults to(2, 2)) — Window size(h, w)for the intermediate window-attention merger.
This is the configuration class to store the configuration of a MiniCPMV4_7Model. It is used to instantiate a Minicpmv4 7 model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the openbmb/MiniCPM-V-4.7
Configuration objects inherit from PreTrainedConfig and can be used to control the model outputs. Read the documentation from PreTrainedConfig for more information.
MiniCPMV4_7VisionPreTrainedModel.
class transformers.MiniCPMV4_7VisionPreTrainedModel
< source >( config: PreTrainedConfig*inputs**kwargs )
A mock value for a dotted path (e.g. torch.float32): attribute access chains,
calls behave as pass-through decorators, repr is the dotted path, and using it
as a base class substitutes a plain-type base (PEP 560 __mro_entries__), so
real subclasses keep a normal metaclass and inspect.signature reads their real __init__ instead of a mock’s.
MiniCPMV4_7VisionModel
forward
< source >( pixel_valuestarget_sizes: typing.Optional[torch.IntTensor] = Noneuse_vit_merger: bool = True**kwargs: Unpack ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Parameters
- pixel_values (`
of shape(batch_size, num_channels, image_size, image_size)) -- The tensors corresponding to the input images. Pixel values can be obtained using [MiniCPMV4_6ImageProcessor](/docs/transformers/main/en/model_doc/minicpmv4_6#transformers.MiniCPMV4_6ImageProcessor). SeeMiniCPMV4_6ImageProcessor.call()` for details (MiniCPMV4_7Processor uses MiniCPMV4_6ImageProcessor for processing images). - target_sizes (
torch.IntTensorof shape(batch_size, 2), optional) — Patch grid sizes(h, w)for computing position embeddings. - use_vit_merger (
bool, optional, defaults toTrue) — Whether to apply the ViT window-attention merger after the encoder.
Returns
BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A BaseModelOutputWithPooling or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
The MiniCPMV4_7VisionModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the
Moduleinstance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.pooler_output (
torch.FloatTensorof shape(batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
MiniCPMV4_7Model
class transformers.MiniCPMV4_7Model
< source >( config: MiniCPMV4_7Config )
Parameters
- config (MiniCPMV4_7Config) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The MiniCPMV4_7 model which consists of a vision backbone and a language model, without a language modeling head.
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_ids: typing.Optional[torch.LongTensor] = Nonepixel_values: typing.Optional[torch.FloatTensor] = Nonetarget_sizes: typing.Optional[torch.IntTensor] = Nonepixel_values_videos: typing.Optional[torch.FloatTensor] = Nonetarget_sizes_videos: typing.Optional[torch.IntTensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Noneposition_ids: typing.Optional[torch.LongTensor] = Nonepast_key_values: list[torch.FloatTensor] | None = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Noneuse_cache: bool | None = Nonedownsample_mode: str | None = Nonemm_token_type_ids: typing.Optional[torch.IntTensor] = None**kwargs: Unpack ) → BaseModelOutputWithPast or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
- pixel_values (
torch.FloatTensor, optional) — Pixel value patches for images, NaViT-packed. - target_sizes (
torch.IntTensor, optional) — Height and width (in patches) for each image. - pixel_values_videos (
torch.FloatTensor, optional) — Pixel value patches for video frames, NaViT-packed. - target_sizes_videos (
torch.IntTensor, optional) — Height and width (in patches) for each video frame. - attention_mask (
torch.Tensorof 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.
- position_ids (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range[0, config.n_positions - 1]. - past_key_values (
list[torch.FloatTensor], optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in thepast_key_valuesreturned by the model at a previous stage of decoding, whenuse_cache=Trueorconfig.use_cache=True.Only Cache instance is allowed as input, see our kv cache guide. If no
past_key_valuesare passed, DynamicCache will be initialized by default.The model will output the same cache format that is fed as input.
If
past_key_valuesare used, the user is expected to input only unprocessedinput_ids(those that don’t have their past key value states given to this model) of shape(batch_size, unprocessed_length)instead of allinput_idsof shape(batch_size, sequence_length). - inputs_embeds (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passinginput_idsyou can choose to directly pass an embedded representation. This is useful if you want more control over how to convertinput_idsindices into associated vectors than the model’s internal embedding lookup matrix. - use_cache (
bool, optional) — If set toTrue,past_key_valueskey value states are returned and can be used to speed up decoding (seepast_key_values). - downsample_mode (
str, optional) —"4x"keeps 4x more visual tokens; default"16x"applies full merge. - mm_token_type_ids (
torch.IntTensorof shape(batch_size, sequence_length), optional) — Indices of input sequence tokens matching each modality. For example text (0), image (1), video (2). Multimodal token type ids can be obtained using AutoProcessor. See ProcessorMixin.call() for details.
Returns
BaseModelOutputWithPast or tuple(torch.FloatTensor)
A BaseModelOutputWithPast or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
The MiniCPMV4_7Model forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the
Moduleinstance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.If
past_key_valuesis used only the last hidden-state of the sequences of shape(batch_size, 1, hidden_size)is output.past_key_values (
Cache, optional, returned whenuse_cache=Trueis passed or whenconfig.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
config.is_encoder_decoder=Truein the cross-attention blocks) that can be used (seepast_key_valuesinput) to speed up sequential decoding.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
get_image_features
< source >( pixel_values: FloatTensortarget_sizes: IntTensordownsample_mode: str | None = None**kwargs: Unpack ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Parameters
- pixel_values (
torch.FloatTensorof shape(batch_size, num_channels, image_size, image_size)) — The tensors corresponding to the input images. Pixel values can be obtained using MiniCPMV4_6ImageProcessor. SeeMiniCPMV4_6ImageProcessor.__call__()for details (MiniCPMV4_7Processor uses MiniCPMV4_6ImageProcessor for processing images). - target_sizes (
torch.IntTensorof shape(num_images, 2)) — Height and width (in patches) of each image. - downsample_mode (
str, optional) — When set to"4x"the intermediatevit_mergeris skipped so that each image keeps4×more visual tokens. Default"16x"mode applies the full merge pipeline.
Returns
BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A BaseModelOutputWithPooling or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
Extract image features: vision encoder, insert merger, then MLP merger.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.pooler_output (
torch.FloatTensorof shape(batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
get_video_features
< source >( pixel_values_videos: FloatTensortarget_sizes_videos: IntTensordownsample_mode: str | None = None**kwargs: Unpack ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Parameters
- pixel_values_videos (
torch.FloatTensorof shape(1, channels, patch_size, seq_len)) — NaViT-packed pixel patches for all video frames. The video processor concatenates every frame’s patches along the last dimension into a single sequence with dim-0 = 1, identical to the image packing format. - target_sizes_videos (
torch.IntTensorof shape(num_patches, 2)) — Height and width (in patches) of each visual unit. - downsample_mode (
str, optional) — When set to"4x"the intermediatevit_mergeris skipped so that each frame keeps4×more visual tokens. Default"16x"mode applies the full merge pipeline.
Returns
BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A BaseModelOutputWithPooling or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
Extract video features: repack frames into NaViT format, then vision encoder + merger.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.pooler_output (
torch.FloatTensorof shape(batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
MiniCPMV4_7ForConditionalGeneration
class transformers.MiniCPMV4_7ForConditionalGeneration
< source >( config: MiniCPMV4_7Config )
Parameters
- config (MiniCPMV4_7Config) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The Minicpmv4 7 Model for token generation conditioned on other modalities (e.g. image-text-to-text generation).
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_ids: typing.Optional[torch.LongTensor] = Nonepixel_values: typing.Optional[torch.FloatTensor] = Nonetarget_sizes: typing.Optional[torch.IntTensor] = Nonepixel_values_videos: typing.Optional[torch.FloatTensor] = Nonetarget_sizes_videos: typing.Optional[torch.IntTensor] = Noneattention_mask: typing.Optional[torch.Tensor] = Noneposition_ids: typing.Optional[torch.LongTensor] = Nonepast_key_values: list[torch.FloatTensor] | None = Noneinputs_embeds: typing.Optional[torch.FloatTensor] = Nonelabels: typing.Optional[torch.LongTensor] = Noneuse_cache: bool | None = Nonedownsample_mode: str | None = Nonemm_token_type_ids: typing.Optional[torch.IntTensor] = Nonelogits_to_keep: typing.Union[int, torch.Tensor] = 0**kwargs: Unpack ) → CausalLMOutputWithPast or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
- pixel_values (
torch.FloatTensor, optional) — Pixel value patches for images, NaViT-packed. - target_sizes (
torch.IntTensor, optional) — Height and width (in patches) for each image. - pixel_values_videos (
torch.FloatTensor, optional) — Pixel value patches for video frames, NaViT-packed. - target_sizes_videos (
torch.IntTensor, optional) — Height and width (in patches) for each video frame. - attention_mask (
torch.Tensorof 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.
- position_ids (
torch.LongTensorof shape(batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range[0, config.n_positions - 1]. - past_key_values (
list[torch.FloatTensor], optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in thepast_key_valuesreturned by the model at a previous stage of decoding, whenuse_cache=Trueorconfig.use_cache=True.Only Cache instance is allowed as input, see our kv cache guide. If no
past_key_valuesare passed, DynamicCache will be initialized by default.The model will output the same cache format that is fed as input.
If
past_key_valuesare used, the user is expected to input only unprocessedinput_ids(those that don’t have their past key value states given to this model) of shape(batch_size, unprocessed_length)instead of allinput_idsof shape(batch_size, sequence_length). - inputs_embeds (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passinginput_idsyou can choose to directly pass an embedded representation. This is useful if you want more control over how to convertinput_idsindices into associated vectors than the model’s internal embedding lookup matrix. - labels (
torch.LongTensorof 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 (seeinput_idsdocstring). Tokens with indices set to-100are ignored (masked), the loss is only computed for the tokens with labels in[0, ..., config.vocab_size]. - use_cache (
bool, optional) — If set toTrue,past_key_valueskey value states are returned and can be used to speed up decoding (seepast_key_values). - downsample_mode (
str, optional) —"4x"keeps 4x more visual tokens; default"16x"applies full merge. - mm_token_type_ids (
torch.IntTensorof shape(batch_size, sequence_length), optional) — Indices of input sequence tokens matching each modality. For example text (0), image (1), video (2). Multimodal token type ids can be obtained using AutoProcessor. See ProcessorMixin.call() for details. - logits_to_keep (
Union[int, torch.Tensor], optional, defaults to0) — If anint, compute logits for the lastlogits_to_keeptokens. If0, calculate logits for allinput_ids(special case). Only last token logits are needed for generation, and calculating them only for that token can save memory, which becomes pretty significant for long sequences or large vocabulary size. If atorch.Tensor, must be 1D corresponding to the indices to keep in the sequence length dimension. This is useful when using packed tensor format (single dimension for batch and sequence length).
Returns
CausalLMOutputWithPast or tuple(torch.FloatTensor)
A CausalLMOutputWithPast or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
The MiniCPMV4_7ForConditionalGeneration forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the
Moduleinstance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.
loss (
torch.FloatTensorof shape(1,), optional, returned whenlabelsis provided) — Language modeling loss (for next-token prediction).logits (
torch.FloatTensorof shape(batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).past_key_values (
Cache, optional, returned whenuse_cache=Trueis passed or whenconfig.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
past_key_valuesinput) to speed up sequential decoding.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
Example:
>>> from PIL import Image
>>> from transformers import AutoProcessor, MiniCPMV4_7ForConditionalGeneration
>>> model = MiniCPMV4_7ForConditionalGeneration.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> processor = AutoProcessor.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> messages = [
... {
... "role": "user", "content": [
... {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
... {"type": "text", "text": "Where is the cat standing?"},
... ]
... },
... ]
>>> inputs = processor.apply_chat_template(
... messages,
... tokenize=True,
... return_dict=True,
... return_tensors="pt",
... add_generation_prompt=True
... )
>>> # Generate
>>> generate_ids = model.generate(**inputs)
>>> processor.batch_decode(generate_ids, skip_special_tokens=True)[0]get_image_features
< source >( *args**kwargs ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Returns
BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A BaseModelOutputWithPooling or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
Extract image features: vision encoder, insert merger, then MLP merger.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.pooler_output (
torch.FloatTensorof shape(batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
Example:
>>> from PIL import Image
>>> from transformers import AutoProcessor, MiniCPMV4_7ForConditionalGeneration
>>> model = MiniCPMV4_7ForConditionalGeneration.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> processor = AutoProcessor.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> messages = [
... {
... "role": "user", "content": [
... {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
... {"type": "text", "text": "Where is the cat standing?"},
... ]
... },
... ]
>>> inputs = processor.apply_chat_template(
... messages,
... tokenize=True,
... return_dict=True,
... return_tensors="pt",
... add_generation_prompt=True
... )
>>> # Generate
>>> generate_ids = model.generate(**inputs)
>>> processor.batch_decode(generate_ids, skip_special_tokens=True)[0]get_video_features
< source >( *args**kwargs ) → BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Returns
BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A BaseModelOutputWithPooling or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MiniCPMV4_7Config) and inputs.
Extract video features: repack frames into NaViT format, then vision encoder + merger.
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.pooler_output (
torch.FloatTensorof shape(batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
Example:
>>> from PIL import Image
>>> from transformers import AutoProcessor, MiniCPMV4_7ForConditionalGeneration
>>> model = MiniCPMV4_7ForConditionalGeneration.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> processor = AutoProcessor.from_pretrained("openbmb/MiniCPM-V-4.7")
>>> messages = [
... {
... "role": "user", "content": [
... {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"},
... {"type": "text", "text": "Where is the cat standing?"},
... ]
... },
... ]
>>> inputs = processor.apply_chat_template(
... messages,
... tokenize=True,
... return_dict=True,
... return_tensors="pt",
... add_generation_prompt=True
... )
>>> # Generate
>>> generate_ids = model.generate(**inputs)
>>> processor.batch_decode(generate_ids, skip_special_tokens=True)[0]MiniCPMV4_7Processor
class transformers.MiniCPMV4_7Processor
< source >( image_processor = Nonevideo_processor = Nonetokenizer = Nonechat_template = None**kwargs )
Parameters
- image_processor (
MiniCPMV4_6ImageProcessor) — The image processor is a required input. - video_processor (
MiniCPMV4_6VideoProcessor) — The video processor is a required input. - tokenizer (
TokenizersBackend) — The tokenizer is a required input. - chat_template (
str) — A Jinja template to convert lists of messages in a chat into a tokenizable string.
Constructs a MiniCPMV4_7Processor which wraps a image processor, a video processor, and a tokenizer into a single processor.
MiniCPMV4_7Processor offers all the functionalities of MiniCPMV4_6ImageProcessor, MiniCPMV4_6VideoProcessor, and TokenizersBackend. See the ~MiniCPMV4_6ImageProcessor, ~MiniCPMV4_6VideoProcessor, and ~TokenizersBackend for more information.
__call__
< source >( images: typing.Union[ForwardRef('PIL.Image.Image'), numpy.ndarray, ForwardRef('torch.Tensor'), list['PIL.Image.Image'], list[numpy.ndarray], list['torch.Tensor'], NoneType] = Nonetext: str | list[str] | list[list[str]] | None = Nonevideos: typing.Union[list['PIL.Image.Image'], numpy.ndarray, ForwardRef('torch.Tensor'), list[numpy.ndarray], list['torch.Tensor'], list[list['PIL.Image.Image']], list[list[numpy.ndarray]], list[list['torch.Tensor']], transformers.video_utils.URL, list[transformers.video_utils.URL], list[list[transformers.video_utils.URL]], transformers.video_utils.Path, list[transformers.video_utils.Path], list[list[transformers.video_utils.Path]], NoneType] = None**kwargs: Unpack )
Parameters
- images (
Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, list[PIL.Image.Image], list[numpy.ndarray], list[torch.Tensor]], optional) — Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If passing in images with pixel values between 0 and 1, setdo_rescale=False. - text (
Union[str, list[str], list[list[str]]], optional) — The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings (pretokenized string). If you pass a pretokenized input, setis_split_into_words=Trueto avoid ambiguity with batched inputs. - videos (
Union[list[PIL.Image.Image], numpy.ndarray, torch.Tensor, list[numpy.ndarray], list[torch.Tensor], list[list[PIL.Image.Image]], list[list[numpy.ndarray]], list[list[torch.Tensor]], ~video_utils.URL, list[~video_utils.URL], list[list[~video_utils.URL]], ~video_utils.Path, list[~video_utils.Path], list[list[~video_utils.Path]]], optional) — Video to preprocess. Expects a single or batch of videos with pixel values ranging from 0 to 255. If passing in videos with pixel values between 0 and 1, setdo_rescale=False. - return_tensors (
stror TensorType, optional) — If set, will return tensors of a particular framework. Acceptable values are:'pt': Return PyTorchtorch.Tensorobjects.'np': Return NumPynp.ndarrayobjects.
- **kwargs (ProcessingKwargs, optional) — Additional processing options for each modality (text, images, videos, audio). Model-specific parameters are listed above; see the TypedDict class for the complete list of supported arguments.