vllm-cpu-amxbf16 0.9.1__cp312-cp312-manylinux_2_17_x86_64.whl

vllm/transformers_utils/configs/exaone.py

@@ -0,0 +1,190 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copied from
+# https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/blob/main/configuration_exaone.py
+# Copyright 2021 The LG AI Research EXAONE Lab. 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.
+"""Exaone model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+EXAONE_PRETRAINED_CONFIG_ARCHIVE_MAP: dict[str, str] = {}
+
+
+class ExaoneConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a :class:
+    `~transformers.ExaoneModel`. It is used to instantiate a GPT Lingvo 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 Exaone
+
+    Configuration objects inherit from {class}`~transformers.PretrainedConfig`
+    and can be used to control the model outputs. Read the documentation from :
+    class:`~transformers.PretrainedConfig` for more information.
+
+    Args:
+        vocab_size ({obj}`int`, `optional`, defaults to 50257):
+            Vocabulary size of the GPT Lingvo model. Defines the number of
+            different tokens that can be represented by the {obj}`inputs_ids`
+            passed when calling {class}`~transformers.ExaoneModel`. Vocabulary
+            size of the model.
+            Defines the different tokens that can be represented by the
+            `inputs_ids` passed to the forward method of :class:
+            `~transformers.EXAONEModel`.
+        hidden_size ({obj}`int`, `optional`, defaults to 2048):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_layers ({obj}`int`, `optional`, defaults to 24):
+            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 decoder.
+        num_key_value_heads (`int`, *optional*):
+            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://arxiv.org/pdf/2305.13245.pdf). If it is not
+            specified, will default to `num_attention_heads`.
+        rotary_pct (`float`, *optional*, defaults to 0.25):
+            percentage of hidden dimensions to allocate to rotary embeddings
+        intermediate_size ({obj}`int`, `optional`, defaults to 8192):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in
+            the Transformer encoder.
+        activation_function ({obj}`str` or {obj}`function`, `optional`,
+        defaults to {obj}`"gelu_new"`):
+            The non-linear activation function (function or string) in the
+            encoder and pooler. If string, {obj}`"gelu"`, {obj}`"relu"`,
+            {obj}`"selu"` and {obj}`"gelu_new"` are supported.
+        embed_dropout ({obj}`float`, `optional`, defaults to 0.0):
+            The dropout probabilitiy for all fully connected layers in the
+            embeddings, encoder, and pooler.
+        attention_dropout ({obj}`float`, `optional`, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings ({obj}`int`, `optional`, defaults to 2048):
+            The maximum sequence length that this model might ever be used with.
+            Typically set this to something large just in case
+            (e.g., 512 or 1024 or 2048).
+        type_vocab_size ({obj}`int`, `optional`, defaults to 2):
+            The vocabulary size of the {obj}`token_type_ids` passed when calling
+            {class}`~transformers.EXAONEModel`.
+        initializer_range ({obj}`float`, `optional`, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for
+            initializing all weight matrices.
+        layer_norm_epsilon ({obj}`float`, `optional`, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        use_cache ({obj}`bool`, `optional`, defaults to {obj}`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``.
+        gradient_checkpointing ({obj}`bool`, `optional`,
+        defaults to {obj}`False`):
+            If True, use gradient checkpointing to save memory at the expense
+            of slower backward pass.
+        Example::
+
+            >>> from transformers import ExoneModel, ExaoneConfig
+
+            >>> # Initializing a EXAONE configuration
+            >>> configuration = ExaoneConfig()
+
+            >>> # Initializing a model from configuration
+            >>> model = ExoneModel(configuration)
+
+            >>> # Accessing the model configuration
+            >>> configuration = model.config
+    """
+
+    model_type = "exaone"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {"num_hidden_layers": "num_layers"}
+
+    def __init__(
+        self,
+        vocab_size=102400,
+        max_position_embeddings=2048,
+        hidden_size=2048,
+        num_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        intermediate_size=None,
+        activation_function="silu",
+        rotary_pct=0.25,
+        resid_dropout=0.0,
+        embed_dropout=0.0,
+        attention_dropout=0.0,
+        layer_norm_epsilon=1e-6,
+        initializer_range=0.02,
+        use_cache=True,
+        bos_token_id=0,
+        eos_token_id=2,
+        tie_word_embeddings=True,
+        **kwargs,
+    ):
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.num_layers = num_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_hidden_layers = num_layers
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        if intermediate_size:
+            self.intermediate_size = intermediate_size
+        else:
+            self.intermediate_size = hidden_size * 4
+        self.activation_function = activation_function
+        self.resid_dropout = resid_dropout
+        self.embed_dropout = embed_dropout
+        self.attention_dropout = attention_dropout
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.use_cache = use_cache
+        self.rotary_pct = rotary_pct
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+
+        self.use_logit_cap = kwargs.pop("use_logit_cap", False)
+        self.ln_no_scale = kwargs.pop("ln_no_scale", False)
+        self.use_gated = kwargs.pop("use_gated", False)
+        self.use_emb_norm = kwargs.pop("use_emb_norm", False)
+        self.use_rotary_pos = kwargs.pop("use_rotary_pos", False)
+        self.rotary_type = kwargs.pop("rotary_type", None)
+        self.scaling_factor = kwargs.pop("scaling_factor", 1)
+        self.use_absolute_pos = kwargs.pop("use_absolute_pos", True)
+        self.use_extra_logit = kwargs.pop("use_extra_logit", True)
+        self.rotary_expand_length = kwargs.pop("rotary_expand_length", None)
+        self.rotary_base = kwargs.pop("rotary_base", 10000.0)
+        self.use_qkv_fuse = kwargs.pop("use_qkv_fuse", False)
+        self.rescale_before_lm_head = kwargs.pop("rescale_before_lm_head",
+                                                 (rotary_pct == 0.25))
+        if self.use_rotary_pos:
+            self.use_absolute_pos = False

vllm/transformers_utils/configs/falcon.py

@@ -0,0 +1,90 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Adapted from
+# https://huggingface.co/tiiuae/falcon-7b/blob/main/configuration_RW.py
+# Copyright 2023 The vLLM team.
+# Copyright 2022 the Big Science Workshop 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.
+"""Falcon configuration"""
+from transformers.configuration_utils import PretrainedConfig
+
+
+class RWConfig(PretrainedConfig):
+    model_type = "falcon"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "num_hidden_layers": "n_layer",
+        "num_attention_heads": "n_head",
+        "num_kv_heads": "n_head_kv",
+    }
+
+    def __init__(
+        self,
+        vocab_size=250880,
+        hidden_size=64,
+        n_layer=2,
+        n_head=8,
+        layer_norm_epsilon=1e-5,
+        initializer_range=0.02,
+        use_cache=True,
+        bos_token_id=1,
+        eos_token_id=2,
+        hidden_dropout=0.0,
+        attention_dropout=0.0,
+        multi_query=True,
+        n_head_kv=None,
+        alibi=False,
+        bias=False,
+        parallel_attn=False,
+        new_decoder_architecture=False,
+        **kwargs,
+    ) -> None:
+        self.vocab_size = vocab_size
+        # Backward compatibility with n_embed kwarg
+        n_embed = kwargs.pop("n_embed", None)
+        self.hidden_size = hidden_size if n_embed is None else n_embed
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.use_cache = use_cache
+        self.hidden_dropout = hidden_dropout
+        self.attention_dropout = attention_dropout
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.multi_query = multi_query
+        self.n_head_kv = 1 if n_head_kv is None else n_head_kv
+        self.alibi = alibi
+        self.bias = bias
+        self.parallel_attn = parallel_attn
+        self.new_decoder_architecture = new_decoder_architecture
+
+        if self.hidden_size == 8192:
+            # Hack for falcon-40b
+            self.new_decoder_architecture = True
+
+        super().__init__(bos_token_id=bos_token_id,
+                         eos_token_id=eos_token_id,
+                         **kwargs)
+
+    @property
+    def head_dim(self):
+        return self.hidden_size // self.n_head
+
+    @property
+    def rotary(self):
+        return not self.alibi

vllm/transformers_utils/configs/h2ovl.py

@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Adapted from
+# https://huggingface.co/h2oai/h2ovl-mississippi-2b/blob/main/configuration_h2ovl_chat.py
+# --------------------------------------------------------
+# H2OVL-Mississippi
+# Copyright (c) 2024 H2O.AI
+# Licensed under Apache 2.0 License [see LICENSE for details]
+# --------------------------------------------------------
+
+from .internvl import InternVLChatConfig
+
+
+class H2OVLChatConfig(InternVLChatConfig):
+    model_type = "h2ovl_chat"

vllm/transformers_utils/configs/internvl.py

@@ -0,0 +1,54 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Adapted from
+# https://huggingface.co/OpenGVLab/InternVL2-1B/blob/main/configuration_internvl_chat.py
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2024 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+from transformers.configuration_utils import PretrainedConfig
+
+
+class InternVLChatConfig(PretrainedConfig):
+    model_type = 'internvl_chat'
+    is_composition = True
+
+    def __init__(self,
+                 vision_config=None,
+                 llm_config=None,
+                 use_backbone_lora=0,
+                 use_llm_lora=0,
+                 select_layer=-1,
+                 force_image_size=None,
+                 downsample_ratio=0.5,
+                 template=None,
+                 dynamic_image_size=False,
+                 use_thumbnail=False,
+                 ps_version='v1',
+                 min_dynamic_patch=1,
+                 max_dynamic_patch=6,
+                 **kwargs):
+        super().__init__(**kwargs)
+
+        if vision_config is None:
+            vision_config = {}
+
+        if llm_config is None:
+            llm_config = {}
+
+        self.vision_config = PretrainedConfig(**vision_config)
+        self.text_config = PretrainedConfig(**llm_config)
+
+        self.use_backbone_lora = use_backbone_lora
+        self.use_llm_lora = use_llm_lora
+        self.select_layer = select_layer
+        self.force_image_size = force_image_size
+        self.downsample_ratio = downsample_ratio
+        self.template = template
+        self.dynamic_image_size = dynamic_image_size
+        self.use_thumbnail = use_thumbnail
+        self.ps_version = ps_version  # pixel shuffle version
+        self.min_dynamic_patch = min_dynamic_patch
+        self.max_dynamic_patch = max_dynamic_patch

vllm/transformers_utils/configs/jais.py

@@ -0,0 +1,238 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright 2023 The OpenAI Team Authors and HuggingFace Inc. team.
+# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
+# Copyright 2023 Cerebras Systems.
+#
+# 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.
+"""JAIS configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class JAISConfig(PretrainedConfig):
+    """
+    This is the configuration class to store the configuration of a
+    [`JAISModel`]. It is used to instantiate a JAIS model according to the
+    specified arguments, defining the model architecture.
+
+    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 50257):
+            Vocabulary size of the JAIS model. Defines the number of different
+            tokens that can be represented by the
+            `inputs_ids` passed when calling [`JAISModel`].
+        n_positions (`int`, *optional*, defaults to 1024):
+            The maximum sequence length that this model might ever be used
+            with. Typically set this to something large just in case
+            (e.g., 512 or 1024 or 2048).
+        n_embd (`int`, *optional*, defaults to 768):
+            Dimensionality of the embeddings and hidden states.
+        n_layer (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        n_head (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the
+            Transformer encoder.
+        n_inner (`int`, *optional*, defaults to None):
+            Dimensionality of the inner feed-forward layers. `None` will set
+            it to 4 times n_embd
+        activation_function (`str`, *optional*, defaults to `"gelu"`):
+            Activation function, to be selected in the list
+            `["relu", "silu", "gelu", "tanh", "gelu_new", "swiglu"]`.
+        resid_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in
+            the embeddings, encoder, and pooler.
+        embd_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the embeddings.
+        attn_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon to use in the layer normalization layers.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for
+            initializing all weight matrices.
+        scale_attn_weights (`bool`, *optional*, defaults to `True`):
+            Scale attention weights by dividing by sqrt(hidden_size)..
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values
+            attentions (not used by all models).
+        scale_attn_by_inverse_layer_idx (`bool`, *optional*,
+            defaults to `False`):
+            Whether to additionally scale attention weights by
+            `1 / layer_idx + 1`.
+        reorder_and_upcast_attn (`bool`, *optional*, defaults to `False`):
+            Whether to scale keys (K) prior to computing attention
+            (dot-product)
+            and upcast attention dot-product/softmax to float() when training
+            with mixed precision.
+        position_embedding_type (`str`, *optional*, defaults to `"learned"`):
+            Positional embedding can be either `"alibi"` or `"learned"`.
+        mup_width_scale (`float`, *optional*, defaults to 1.0):
+            muP parameter to scale learning rate and initializers. Calculated
+            as (`d_model,0 / d_model`), where
+            `d_model` is the model's width and `d_model,0` is the proxy
+            model's width.
+        mup_embeddings_scale (`float`, *optional*, defaults to 1.0):
+            muP parameter to scale token and position embeddings.
+        mup_output_alpha (`float`, *optional*, defaults to 1.0):
+            muP parameter to scale output logits
+            (`output_logits_scale = mup_output_alpha * mup_width_scale`).
+        mup_scale_qk_dot_by_d (`bool`, *optional*, defaults to `False`):
+            Scale attention weights by dividing by hidden_size instead of
+            sqrt(hidden_size). Need to set scale_attn_weights to `True` as
+            well.
+        alibi_scaling (`dict`, *optional*):
+            Dictionary containing the scaling configuration for ALiBi
+            embeddings. Currently only supports linear
+            scaling strategy. Can specify either the scaling `factor` (must be
+            a float greater than 1) for fixed scaling
+            or `train_seq_len` for dynamic scaling on input samples with
+            sequence length > `train_seq_len`. The expected
+            formats are `{"type": strategy name, "factor": scaling factor}` or
+            `{"type": strategy name,
+            "train_seq_len": training sequence length}`.
+        architectures (`list`, *optional*, defaults to ['JAISLMHeadModel']):
+            architecture names for Jais.
+
+    Example:
+
+    ```python
+    >>> from transformers import JAISConfig, JAISModel
+
+    >>> # Initializing a JAIS configuration
+    >>> configuration = JAISConfig()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = JAISModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "jais"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "hidden_size": "n_embd",
+        "max_position_embeddings": "n_positions",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+    }
+
+    def __init__(
+        self,
+        vocab_size=50257,
+        n_positions=1024,
+        n_embd=768,
+        n_layer=12,
+        n_head=12,
+        n_inner=None,
+        activation_function="gelu_new",
+        resid_pdrop=0.1,
+        embd_pdrop=0.1,
+        attn_pdrop=0.1,
+        layer_norm_epsilon=1e-5,
+        initializer_range=0.02,
+        scale_attn_weights=True,
+        use_cache=True,
+        bos_token_id=50256,
+        eos_token_id=50256,
+        scale_attn_by_inverse_layer_idx=False,
+        reorder_and_upcast_attn=False,
+        position_embedding_type="learned",
+        mup_width_scale=1.0,
+        mup_embeddings_scale=1.0,
+        mup_output_alpha=1.0,
+        mup_scale_qk_dot_by_d=False,
+        alibi_scaling=None,
+        architectures=None,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.n_inner = n_inner
+        self.activation_function = activation_function
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attn_pdrop = attn_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.scale_attn_weights = scale_attn_weights
+        self.use_cache = use_cache
+        self.scale_attn_by_inverse_layer_idx = scale_attn_by_inverse_layer_idx
+        self.reorder_and_upcast_attn = reorder_and_upcast_attn
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+
+        self.position_embedding_type = position_embedding_type
+        self.mup_width_scale = mup_width_scale
+        self.mup_embeddings_scale = mup_embeddings_scale
+        self.mup_output_alpha = mup_output_alpha
+        self.mup_scale_qk_dot_by_d = mup_scale_qk_dot_by_d
+
+        self.alibi_scaling = alibi_scaling
+        self._alibi_scaling_validation()
+        if architectures is None:
+            architectures = ["JAISLMHeadModel"]
+
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            architectures=architectures,
+            **kwargs,
+        )
+
+    def _alibi_scaling_validation(self):
+        """
+        Validate the `alibi_scaling` configuration.
+        """
+        if self.alibi_scaling is None:
+            return
+
+        if (not isinstance(self.alibi_scaling, dict)
+                or len(self.alibi_scaling) != 2):
+            raise ValueError(
+                "`alibi_scaling` must be a dictionary with two fields, "
+                "`type` and `factor` or `type` and `train_seq_len`, "
+                f"got {self.alibi_scaling}")
+        alibi_scaling_type = self.alibi_scaling.get("type", None)
+        alibi_scaling_factor = self.alibi_scaling.get("factor", None)
+        alibi_dynamic_scaling = self.alibi_scaling.get("train_seq_len", None)
+        if alibi_scaling_type is None or alibi_scaling_type != "linear":
+            raise ValueError(f"`alibi_scaling`'s type field must be 'linear', "
+                             f"got {alibi_scaling_type}")
+        if (alibi_scaling_factor is not None
+                and not isinstance(alibi_scaling_factor, float)
+                or (alibi_scaling_factor is not None
+                    and alibi_scaling_factor <= 1.0)):
+            raise ValueError(
+                f"`alibi_scaling`'s factor field must be a float > 1.0, "
+                f"got {alibi_scaling_factor}")
+        if (alibi_dynamic_scaling is not None
+                and not isinstance(alibi_dynamic_scaling, int)
+                or (alibi_dynamic_scaling is not None
+                    and alibi_dynamic_scaling <= 1)):
+            raise ValueError(
+                f"`alibi_scaling`'s `train_seq_len` field must be an "
+                f"integer > 1, got {alibi_dynamic_scaling}")

vllm/transformers_utils/configs/kimi_vl.py

@@ -0,0 +1,37 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# Adapted from https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct/blob/main/configuration_kimi_vl.py
+from typing import Optional, Union
+
+from transformers.configuration_utils import PretrainedConfig
+
+from vllm.transformers_utils.configs.deepseek_vl2 import DeepseekV2Config
+from vllm.transformers_utils.configs.moonvit import MoonViTConfig
+
+
+class KimiVLConfig(PretrainedConfig):
+    model_type = "kimi_vl"
+
+    def __init__(self,
+                 vision_config: Optional[Union[dict, MoonViTConfig]] = None,
+                 text_config: Optional[Union[dict, DeepseekV2Config]] = None,
+                 ignore_index: int = -100,
+                 media_placeholder_token_id: int = 163605,
+                 pad_token_id: int = 0,
+                 **kwargs):
+        if vision_config is None:
+            vision_config = MoonViTConfig()
+        elif isinstance(vision_config, dict):
+            vision_config = MoonViTConfig(**vision_config)
+        self.vision_config = vision_config
+
+        if text_config is None:
+            text_config = DeepseekV2Config()
+        elif isinstance(text_config, dict):
+            text_config = DeepseekV2Config(**text_config)
+        self.text_config = text_config
+
+        self.ignore_index = ignore_index
+        self.media_placeholder_token_id = media_placeholder_token_id
+
+        super().__init__(pad_token_id=pad_token_id, **kwargs)