vllm.model_executor.models.chameleon
ChameleonAttention
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 |
|
attn
instance-attribute
¶
attn = Attention(
num_heads,
head_dim,
scaling,
num_kv_heads=num_kv_heads,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
o_proj
instance-attribute
¶
o_proj = RowParallelLinear(
input_size=total_num_heads * head_dim,
output_size=hidden_size,
bias=bias,
quant_config=quant_config,
)
qkv_proj
instance-attribute
¶
qkv_proj = QKVParallelLinear(
hidden_size=hidden_size,
head_size=head_dim,
total_num_heads=total_num_heads,
total_num_kv_heads=total_num_kv_heads,
bias=bias,
quant_config=quant_config,
)
rotary_emb
instance-attribute
¶
rotary_emb = get_rope(
head_dim,
rotary_dim=head_dim,
max_position=max_position_embeddings,
base=rope_theta,
rope_scaling=rope_scaling,
)
__init__
¶
__init__(
hidden_size: int,
num_heads: int,
num_kv_heads: int,
rope_theta: float = 10000,
rope_scaling: Optional[dict[str, Any]] = None,
max_position_embeddings: int = 4096,
quant_config: Optional[QuantizationConfig] = None,
bias: bool = False,
cache_config: Optional[CacheConfig] = None,
prefix: str = "",
) -> None
Source code in vllm/model_executor/models/chameleon.py
_apply_qk_norm
¶
Source code in vllm/model_executor/models/chameleon.py
forward
¶
Source code in vllm/model_executor/models/chameleon.py
ChameleonDecoderLayer
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
mlp
instance-attribute
¶
mlp = ChameleonMLP(
hidden_size=hidden_size,
intermediate_size=intermediate_size,
hidden_act=hidden_act,
quant_config=quant_config,
bias=getattr(config, "mlp_bias", False),
)
post_attention_layernorm
instance-attribute
¶
post_attention_layernorm = RMSNorm(
hidden_size, eps=rms_norm_eps
)
self_attn
instance-attribute
¶
self_attn = ChameleonAttention(
hidden_size=hidden_size,
num_heads=num_attention_heads,
num_kv_heads=getattr(
config, "num_key_value_heads", num_attention_heads
),
rope_theta=rope_theta,
rope_scaling=rope_scaling,
max_position_embeddings=max_position_embeddings,
quant_config=quant_config,
bias=False,
cache_config=cache_config,
prefix=f"{prefix}.self_attn",
)
__init__
¶
__init__(
config: ChameleonConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(
positions: Tensor,
hidden_states: Tensor,
residual: Optional[Tensor],
) -> tuple[Tensor, Optional[Tensor]]
Source code in vllm/model_executor/models/chameleon.py
ChameleonDummyInputsBuilder
¶
Bases: BaseDummyInputsBuilder[ChameleonProcessingInfo]
Source code in vllm/model_executor/models/chameleon.py
get_dummy_mm_data
¶
get_dummy_mm_data(
seq_len: int, mm_counts: Mapping[str, int]
) -> MultiModalDataDict
Source code in vllm/model_executor/models/chameleon.py
get_dummy_text
¶
ChameleonForConditionalGeneration
¶
Bases: Module
, SupportsMultiModal
, SupportsPP
, SupportsQuant
Source code in vllm/model_executor/models/chameleon.py
925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 |
|
logits_processor
instance-attribute
¶
logits_processor = LogitsProcessor(
unpadded_vocab_size, vocab_size, logit_scale
)
make_empty_intermediate_tensors
instance-attribute
¶
model
instance-attribute
¶
model = ChameleonModel(
vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"),
)
packed_modules_mapping
class-attribute
instance-attribute
¶
packed_modules_mapping = {
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
"gate_up_proj": ["gate_proj", "up_proj"],
}
__init__
¶
__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/chameleon.py
_parse_and_validate_image_input
¶
_parse_and_validate_image_input(
**kwargs: object,
) -> Optional[ChameleonImagePixelInputs]
Source code in vllm/model_executor/models/chameleon.py
_validate_pixel_values
¶
Source code in vllm/model_executor/models/chameleon.py
compute_logits
¶
compute_logits(
hidden_states: Tensor,
sampling_metadata: SamplingMetadata,
) -> Optional[Tensor]
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(
input_ids: Tensor,
positions: Tensor,
intermediate_tensors: Optional[
IntermediateTensors
] = None,
inputs_embeds: Optional[Tensor] = None,
**kwargs,
) -> Union[Tensor, IntermediateTensors]
Source code in vllm/model_executor/models/chameleon.py
get_input_embeddings
¶
get_input_embeddings(
input_ids: Tensor,
multimodal_embeddings: Optional[
MultiModalEmbeddings
] = None,
) -> Tensor
Source code in vllm/model_executor/models/chameleon.py
get_multimodal_embeddings
¶
get_multimodal_embeddings(
**kwargs: object,
) -> MultiModalEmbeddings
Source code in vllm/model_executor/models/chameleon.py
get_placeholder_str
classmethod
¶
load_weights
¶
Source code in vllm/model_executor/models/chameleon.py
1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 |
|
ChameleonImageVocabularyMapping
¶
A class for mapping discrete image tokens from VQGAN to BPE tokens.
Source code in vllm/model_executor/models/chameleon.py
ChameleonLayerNorm
¶
Bases: LayerNorm
Source code in vllm/model_executor/models/chameleon.py
ChameleonMLP
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
down_proj
instance-attribute
¶
down_proj = RowParallelLinear(
input_size=intermediate_size,
output_size=hidden_size,
bias=bias,
quant_config=quant_config,
)
gate_up_proj
instance-attribute
¶
gate_up_proj = MergedColumnParallelLinear(
input_size=hidden_size,
output_sizes=[intermediate_size] * 2,
bias=bias,
quant_config=quant_config,
)
__init__
¶
__init__(
hidden_size: int,
intermediate_size: int,
hidden_act: str,
quant_config: Optional[QuantizationConfig] = None,
bias: bool = False,
) -> None
Source code in vllm/model_executor/models/chameleon.py
ChameleonModel
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 |
|
make_empty_intermediate_tensors
instance-attribute
¶
make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], hidden_size
)
)
vocabulary_mapping
instance-attribute
¶
vocabulary_mapping = ChameleonImageVocabularyMapping(
vocabulary_map
)
__init__
¶
__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(
input_ids: Optional[Tensor],
positions: Tensor,
intermediate_tensors: Optional[IntermediateTensors],
inputs_embeds: Optional[Tensor] = None,
) -> Union[Tensor, IntermediateTensors]
Source code in vllm/model_executor/models/chameleon.py
get_image_tokens
¶
Tokenizes images into discrete tokens with VQGAN module. Converts obtained image tokens into BPE tokens and wraps with "boi" and "eoi" special tokens.
Source code in vllm/model_executor/models/chameleon.py
ChameleonMultiModalProcessor
¶
Bases: BaseMultiModalProcessor[ChameleonProcessingInfo]
Source code in vllm/model_executor/models/chameleon.py
_apply_hf_processor_tokens_only
¶
Source code in vllm/model_executor/models/chameleon.py
_call_hf_processor
¶
_call_hf_processor(
prompt: str,
mm_data: Mapping[str, object],
mm_kwargs: Mapping[str, object],
tok_kwargs: Mapping[str, object],
) -> BatchFeature
Source code in vllm/model_executor/models/chameleon.py
_get_mm_fields_config
¶
_get_prompt_updates
¶
_get_prompt_updates(
mm_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
out_mm_kwargs: MultiModalKwargs,
) -> Sequence[PromptUpdate]
Source code in vllm/model_executor/models/chameleon.py
ChameleonProcessingInfo
¶
Bases: BaseProcessingInfo
Source code in vllm/model_executor/models/chameleon.py
ChameleonSwinDecoderLayer
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
mlp
instance-attribute
¶
mlp = ChameleonMLP(
hidden_size=hidden_size,
intermediate_size=intermediate_size,
hidden_act=hidden_act,
quant_config=quant_config,
bias=getattr(config, "mlp_bias", False),
)
post_attention_layernorm
instance-attribute
¶
post_attention_layernorm = RMSNorm(
hidden_size, eps=rms_norm_eps
)
self_attn
instance-attribute
¶
self_attn = ChameleonAttention(
hidden_size=hidden_size,
num_heads=num_attention_heads,
num_kv_heads=getattr(
config, "num_key_value_heads", num_attention_heads
),
rope_theta=rope_theta,
rope_scaling=rope_scaling,
max_position_embeddings=max_position_embeddings,
quant_config=quant_config,
bias=False,
cache_config=cache_config,
prefix=f"{prefix}.self_attn",
)
__init__
¶
__init__(
config: ChameleonConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(
positions: Tensor,
hidden_states: Tensor,
residual: Optional[Tensor],
) -> tuple[Tensor, Tensor]
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAE
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
__init__
¶
Source code in vllm/model_executor/models/chameleon.py
encode
¶
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAEEncoder
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 |
|
conv_in
instance-attribute
¶
conv_in = Conv2d(
in_channels,
base_channels,
kernel_size=3,
stride=1,
padding=1,
)
conv_out
instance-attribute
¶
conv_out = Conv2d(
block_in,
2 * latent_channels
if double_latent
else latent_channels,
kernel_size=3,
stride=1,
padding=1,
)
norm_out
instance-attribute
¶
norm_out = GroupNorm(
num_groups=32,
num_channels=block_in,
eps=1e-06,
affine=True,
)
__init__
¶
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(pixel_values: Tensor)
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAEEncoderAttnBlock
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
norm
instance-attribute
¶
norm = GroupNorm(
num_groups=32,
num_channels=in_channels,
eps=1e-06,
affine=True,
)
proj_out
instance-attribute
¶
proj_out = Conv2d(
in_channels,
in_channels,
kernel_size=1,
stride=1,
padding=0,
)
__init__
¶
__init__(in_channels: int)
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(hidden_states: Tensor)
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAEEncoderConvDownsample
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAEEncoderResnetBlock
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
conv1
instance-attribute
¶
conv1 = Conv2d(
in_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1,
)
conv2
instance-attribute
¶
conv2 = Conv2d(
out_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1,
)
conv_shortcut
instance-attribute
¶
conv_shortcut = Conv2d(
in_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1,
)
nin_shortcut
instance-attribute
¶
nin_shortcut = Conv2d(
in_channels,
out_channels,
kernel_size=1,
stride=1,
padding=0,
)
norm1
instance-attribute
¶
norm1 = GroupNorm(
num_groups=32,
num_channels=in_channels,
eps=1e-06,
affine=True,
)
norm2
instance-attribute
¶
norm2 = GroupNorm(
num_groups=32,
num_channels=out_channels,
eps=1e-06,
affine=True,
)
out_channels
instance-attribute
¶
__init__
¶
__init__(
config: ChameleonVQVAEConfig,
in_channels: int,
out_channels=None,
conv_shortcut=False,
)
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(hidden_states: Tensor)
Source code in vllm/model_executor/models/chameleon.py
ChameleonVQVAEVectorQuantizer
¶
Bases: Module
Source code in vllm/model_executor/models/chameleon.py
__init__
¶
Source code in vllm/model_executor/models/chameleon.py
forward
¶
forward(hidden_state: Tensor)