EngineX-Mthreads/enginex-mthreads-vllm
4
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init

Browse Source
This commit is contained in:
luopingyi
2026-01-09 13:34:11 +08:00
parent dfa6476b58
commit b2ef04d792
538 changed files with 105693 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
vllm/transformers_utils/configs/__init__.py Normal file
View File

@@ -0,0 +1,16 @@
from vllm.transformers_utils.configs.chatglm import ChatGLMConfig
from vllm.transformers_utils.configs.dbrx import DbrxConfig
# RWConfig is for the original tiiuae/falcon-40b(-instruct) and
# tiiuae/falcon-7b(-instruct) models. Newer Falcon models will use the
# `FalconConfig` class from the official HuggingFace transformers library.
from vllm.transformers_utils.configs.falcon import RWConfig
from vllm.transformers_utils.configs.jais import JAISConfig
from vllm.transformers_utils.configs.mpt import MPTConfig
__all__ = [
"ChatGLMConfig",
"DbrxConfig",
"MPTConfig",
"RWConfig",
"JAISConfig",
]

68
vllm/transformers_utils/configs/chatglm.py Normal file
View File

@@ -0,0 +1,68 @@
# coding=utf-8
# Adapted from
# https://github.com/THUDM/ChatGLM2-6B
from transformers import PretrainedConfig
class ChatGLMConfig(PretrainedConfig):
model_type = "chatglm"
attribute_map = {
"num_hidden_layers": "num_layers",
"n_head_kv": "multi_query_group_num",
}
def __init__(self,
num_layers=28,
padded_vocab_size=65024,
hidden_size=4096,
ffn_hidden_size=13696,
kv_channels=128,
num_attention_heads=32,
seq_length=2048,
hidden_dropout=0.0,
attention_dropout=0.0,
layernorm_epsilon=1e-5,
rmsnorm=True,
apply_residual_connection_post_layernorm=False,
post_layer_norm=True,
add_bias_linear=False,
add_qkv_bias=False,
interleaved_qkv=False,
bias_dropout_fusion=True,
multi_query_attention=False,
multi_query_group_num=1,
apply_query_key_layer_scaling=True,
attention_softmax_in_fp32=True,
fp32_residual_connection=False,
quantization_bit=0,
pre_seq_len=None,
prefix_projection=False,
**kwargs):
self.num_layers = num_layers
self.vocab_size = padded_vocab_size
self.padded_vocab_size = padded_vocab_size
self.hidden_size = hidden_size
self.ffn_hidden_size = ffn_hidden_size
self.kv_channels = kv_channels
self.num_attention_heads = num_attention_heads
self.seq_length = seq_length
self.hidden_dropout = hidden_dropout
self.attention_dropout = attention_dropout
self.layernorm_epsilon = layernorm_epsilon
self.rmsnorm = rmsnorm
self.apply_residual_connection_post_layernorm = (
apply_residual_connection_post_layernorm)
self.post_layer_norm = post_layer_norm
self.add_bias_linear = add_bias_linear
self.add_qkv_bias = add_qkv_bias
self.bias_dropout_fusion = bias_dropout_fusion
self.multi_query_attention = multi_query_attention
self.multi_query_group_num = multi_query_group_num
self.apply_query_key_layer_scaling = apply_query_key_layer_scaling
self.attention_softmax_in_fp32 = attention_softmax_in_fp32
self.fp32_residual_connection = fp32_residual_connection
self.quantization_bit = quantization_bit
self.pre_seq_len = pre_seq_len
self.prefix_projection = prefix_projection
self.interleaved_qkv = interleaved_qkv
super().__init__(**kwargs)

278
vllm/transformers_utils/configs/dbrx.py Normal file
View File

@@ -0,0 +1,278 @@
# yapf: disable
# ruff: noqa: E501
# coding=utf-8
# Copied from
# https://huggingface.co/databricks/dbrx-base/blob/main/configuration_dbrx.py
"""Dbrx configuration."""
from typing import Any, Optional
from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging
logger = logging.get_logger(__name__)
DBRX_PRETRAINED_CONFIG_ARCHIVE_MAP = {} # type: ignore
class DbrxAttentionConfig(PretrainedConfig):
"""Configuration class for Dbrx Attention.
[`DbrxAttention`] class. It is used to instantiate attention layers
according to the specified arguments, defining the layers architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.
Args:
attn_pdrop (`float`, *optional*, defaults to 0.0):
The dropout probability for the attention layers.
clip_qkv (`float`, *optional*, defaults to None):
If not `None`, clip the queries, keys, and values in the attention layer to this value.
kv_n_heads (Optional[int]): For grouped_query_attention only, allow user to specify number of kv heads.
rope_theta (float): The base frequency for rope.
"""
def __init__(
self,
attn_pdrop: float = 0,
clip_qkv: Optional[float] = None,
kv_n_heads: int = 1,
rope_theta: float = 10000.0,
**kwargs: Any,
):
super().__init__(**kwargs)
self.attn_pdrop = attn_pdrop
self.clip_qkv = clip_qkv
self.kv_n_heads = kv_n_heads
self.rope_theta = rope_theta
for k in ["model_type"]:
if k in kwargs:
kwargs.pop(k)
if len(kwargs) != 0:
raise ValueError(f"Found unknown {kwargs=}")
@classmethod
def from_pretrained(
cls, pretrained_model_name_or_path: str, **kwargs: Any
) -> "PretrainedConfig":
cls._set_token_in_kwargs(kwargs)
config_dict, kwargs = cls.get_config_dict(
pretrained_model_name_or_path, **kwargs
)
if config_dict.get("model_type") == "dbrx":
config_dict = config_dict["attn_config"]
if (
"model_type" in config_dict
and hasattr(cls, "model_type")
and config_dict["model_type"] != cls.model_type
):
logger.warning(
"You are using a model of type %s to instantiate a model of "
"type %s. This is not supported for all configurations of "
"models and can yield errors.",
config_dict["model_type"], cls.model_type)
return cls.from_dict(config_dict, **kwargs)
class DbrxFFNConfig(PretrainedConfig):
"""Configuration class for Dbrx FFN.
[`DbrxFFN`] class. It is used to instantiate feedforward layers according to
the specified arguments, defining the layers architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.
Args:
ffn_act_fn (dict, optional): A dict specifying activation function for the FFN.
The dict should have a key 'name' with the value being the name of
the activation function along with any additional keyword arguments.
ffn_hidden_size (int, optional): The hidden size of the feedforward network.
moe_num_experts (int, optional): The number of experts in the mixture of experts layer.
moe_top_k (int, optional): The number of experts to use in the mixture of experts layer.
moe_jitter_eps (float, optional): The jitter epsilon for the mixture of experts layer.
moe_loss_weight (float, optional): The loss weight for the mixture of experts layer.
moe_normalize_expert_weights (float, optional): The normalization factor for the expert weights.
uniform_expert_assignment (bool, optional): Whether to use uniform expert assignment.
This should only be used for benchmarking purposes.
"""
def __init__(
self,
ffn_act_fn: Optional[dict] = None,
ffn_hidden_size: int = 3584,
moe_num_experts: int = 4,
moe_top_k: int = 1,
moe_jitter_eps: Optional[float] = None,
moe_loss_weight: float = 0.01,
moe_normalize_expert_weights: Optional[float] = 1,
uniform_expert_assignment: bool = False,
**kwargs: Any,
):
super().__init__()
if ffn_act_fn is None:
ffn_act_fn = {"name": "silu"}
self.ffn_act_fn = ffn_act_fn
self.ffn_hidden_size = ffn_hidden_size
self.moe_num_experts = moe_num_experts
self.moe_top_k = moe_top_k
self.moe_jitter_eps = moe_jitter_eps
self.moe_loss_weight = moe_loss_weight
self.moe_normalize_expert_weights = moe_normalize_expert_weights
self.uniform_expert_assignment = uniform_expert_assignment
for k in ["model_type"]:
if k in kwargs:
kwargs.pop(k)
if len(kwargs) != 0:
raise ValueError(f"Found unknown {kwargs=}")
@classmethod
def from_pretrained(
cls, pretrained_model_name_or_path: str, **kwargs: Any
) -> "PretrainedConfig":
cls._set_token_in_kwargs(kwargs)
config_dict, kwargs = cls.get_config_dict(
pretrained_model_name_or_path, **kwargs
)
if config_dict.get("model_type") == "dbrx":
config_dict = config_dict["ffn_config"]
if (
"model_type" in config_dict
and hasattr(cls, "model_type")
and config_dict["model_type"] != cls.model_type
):
logger.warning(
"You are using a model of type %s to instantiate a model of "
"type %s. This is not supported for all "
"configurations of models and can yield errors.", config_dict["model_type"], cls.model_type)
return cls.from_dict(config_dict, **kwargs)
class DbrxConfig(PretrainedConfig):
"""Configuration class for Dbrx.
[`DbrxModel`]. It is used to instantiate a Dbrx 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:
d_model (`int`, *optional*, defaults to 6144):
Dimensionality of the embeddings and hidden states.
n_heads (`int`, *optional*, defaults to 48):
Number of attention heads for each attention layer in the Transformer encoder.
n_layers (`int`, *optional*, defaults to 40):
Number of hidden layers in the Transformer encoder.
max_seq_len (`int`, *optional*, defaults to 32768):
The maximum sequence length of the model.
vocab_size (`int`, *optional*, defaults to 100352):
Vocabulary size of the Dbrx model. Defines the maximum number of different tokens that can be represented by
the `inputs_ids` passed when calling [`DbrxModel`].
resid_pdrop (`float`, *optional*, defaults to 0.0):
The dropout probability applied to the attention output before combining with residual.
emb_pdrop (`float`, *optional*, defaults to 0.0):
The dropout probability for the embedding layer.
attn_config (`dict`, *optional*):
A dictionary used to configure the model's attention module.
ffn_config (`dict`, *optional*):
A dictionary used to configure the model's FFN module.
use_cache (`bool`, *optional*, defaults to `False`):
Whether or not the model should return the last key/values attentions (not used by all models).
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
output_router_logits (`bool`, *optional*, defaults to `False`):
Whether or not the router logits should be returned by the model. Enabling this will also
allow the model to output the auxiliary loss. See [here]() for more details
router_aux_loss_coef (`float`, *optional*, defaults to 0.001):
The aux loss factor for the total loss.
Example:
```python
>>> from transformers import DbrxConfig, DbrxModel
>>> # Initializing a Dbrx configuration
>>> configuration = DbrxConfig()
>>> # Initializing a model (with random weights) from the configuration
>>> model = DbrxModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```
"""
model_type = "dbrx"
attribute_map = {
"num_attention_heads": "n_heads",
"hidden_size": "d_model",
"num_hidden_layers": "n_layers",
"max_position_embeddings": "max_seq_len",
}
def __init__(
self,
d_model: int = 2048,
n_heads: int = 16,
n_layers: int = 24,
max_seq_len: int = 2048,
vocab_size: int = 32000,
resid_pdrop: float = 0.0,
emb_pdrop: float = 0.0,
attn_config: Optional[DbrxAttentionConfig] = None,
ffn_config: Optional[DbrxFFNConfig] = None,
use_cache: bool = True,
initializer_range: float = 0.02,
output_router_logits: bool = False,
router_aux_loss_coef: float = 0.05,
**kwargs: Any,
):
if attn_config is None:
self.attn_config = DbrxAttentionConfig()
elif isinstance(attn_config, dict):
self.attn_config = DbrxAttentionConfig(**attn_config)
else:
self.attn_config = attn_config
if ffn_config is None:
self.ffn_config = DbrxFFNConfig()
elif isinstance(ffn_config, dict):
self.ffn_config = DbrxFFNConfig(**ffn_config)
else:
self.ffn_config = ffn_config
self.d_model = d_model
self.n_heads = n_heads
self.n_layers = n_layers
self.max_seq_len = max_seq_len
self.vocab_size = vocab_size
self.resid_pdrop = resid_pdrop
self.emb_pdrop = emb_pdrop
self.use_cache = use_cache
self.initializer_range = initializer_range
self.output_router_logits = output_router_logits
self.router_aux_loss_coef = router_aux_loss_coef
tie_word_embeddings = kwargs.pop("tie_word_embeddings", False)
if tie_word_embeddings:
raise ValueError(
"tie_word_embeddings is not supported for Dbrx models."
)
super().__init__(
tie_word_embeddings=tie_word_embeddings,
**kwargs,
)

87
vllm/transformers_utils/configs/falcon.py Normal file
View File

@@ -0,0 +1,87 @@
# 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

237
vllm/transformers_utils/configs/jais.py Normal file
View File

@@ -0,0 +1,237 @@
# coding=utf-8
# Copyright 2023 The OpenAI Team Authors and HuggingFace Inc. team.
# Copyright (c) 2024 - 2024 Moore Threads Technology Co., Ltd("Moore Threads"). All rights reserved.
# 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}")

178
vllm/transformers_utils/configs/mpt.py Normal file
View File

@@ -0,0 +1,178 @@
# coding=utf-8
# Copied from
# https://huggingface.co/mosaicml/mpt-7b/blob/main/configuration_mpt.py
"""A HuggingFace-style model configuration."""
import warnings
from typing import Any, Dict, Optional, Union
from transformers import PretrainedConfig
attn_config_defaults: Dict = {
'attn_type': 'multihead_attention',
'attn_pdrop': 0.0,
'attn_impl': 'triton',
'qk_ln': False,
'clip_qkv': None,
'softmax_scale': None,
'prefix_lm': False,
'attn_uses_sequence_id': False,
'alibi': False,
'alibi_bias_max': 8
}
ffn_config_defaults: Dict = {'ffn_type': 'mptmlp'}
init_config_defaults: Dict = {
'name': 'kaiming_normal_',
'fan_mode': 'fan_in',
'init_nonlinearity': 'relu',
'init_div_is_residual': True,
'emb_init_std': None,
'emb_init_uniform_lim': None,
'init_std': None,
'init_gain': 0.0
}
class MPTConfig(PretrainedConfig):
model_type = 'mpt'
attribute_map = {
'num_attention_heads': 'n_heads',
'hidden_size': 'd_model',
'num_hidden_layers': 'n_layers',
}
# pylint: disable=dangerous-default-value
def __init__(self,
d_model: int = 2048,
n_heads: int = 16,
n_layers: int = 24,
expansion_ratio: int = 4,
max_seq_len: int = 2048,
vocab_size: int = 50368,
resid_pdrop: float = 0.0,
emb_pdrop: float = 0.0,
learned_pos_emb: bool = True,
attn_config: Dict = attn_config_defaults,
ffn_config: Dict = ffn_config_defaults,
init_device: str = 'cpu',
logit_scale: Optional[Union[float, str]] = None,
no_bias: bool = False,
embedding_fraction: float = 1.0,
norm_type: str = 'low_precision_layernorm',
use_cache: bool = False,
init_config: Dict = init_config_defaults,
fc_type: str = 'torch',
verbose: Optional[int] = None,
**kwargs: Any):
self.d_model = d_model
self.n_heads = n_heads
self.n_layers = n_layers
self.expansion_ratio = expansion_ratio
self.max_seq_len = max_seq_len
self.vocab_size = vocab_size
self.resid_pdrop = resid_pdrop
self.emb_pdrop = emb_pdrop
self.learned_pos_emb = learned_pos_emb
self.attn_config = attn_config
self.ffn_config = ffn_config
self.init_device = init_device
self.logit_scale = logit_scale
self.no_bias = no_bias
self.embedding_fraction = embedding_fraction
self.norm_type = norm_type
self.use_cache = use_cache
self.init_config = init_config
self.fc_type = fc_type
if verbose is not None:
warnings.warn(DeprecationWarning(
'verbose argument for MPTConfig is now ignored and '
'will be removed. Use python_log_level instead.'),
stacklevel=2)
if 'name' in kwargs:
del kwargs['name']
if 'loss_fn' in kwargs:
del kwargs['loss_fn']
if self.attn_config.get('alibi', False):
self.learned_pos_emb = False
warnings.warn(
f'alibi is turned on, setting `learned_pos_emb` '
f'to {self.learned_pos_emb}`',
stacklevel=2)
super().__init__(**kwargs)
self._validate_config()
def _set_config_defaults(
self, config: Dict[str, Any],
config_defaults: Dict[str, Any]) -> Dict[str, Any]:
for (k, v) in config_defaults.items():
if k not in config:
config[k] = v
return config
def _validate_config(self) -> None:
self.attn_config = self._set_config_defaults(self.attn_config,
attn_config_defaults)
self.ffn_config = self._set_config_defaults(self.ffn_config,
ffn_config_defaults)
self.init_config = self._set_config_defaults(self.init_config,
init_config_defaults)
if self.d_model % self.n_heads != 0:
raise ValueError('d_model must be divisible by n_heads')
if any((
prob < 0 or prob > 1 for prob in
[self.attn_config['attn_pdrop'], self.resid_pdrop, self.emb_pdrop]
)):
raise ValueError(
"self.attn_config['attn_pdrop'], resid_pdrop, emb_pdrop are "
"probabilities and must be between 0 and 1")
if self.attn_config['attn_impl'] not in ['torch', 'flash', 'triton']:
raise ValueError(
f"Unknown attn_impl={self.attn_config['attn_impl']}")
if self.attn_config['prefix_lm'] and self.attn_config[
'attn_impl'] not in ['torch', 'triton']:
raise NotImplementedError(
'prefix_lm only implemented with torch and triton attention.')
if self.attn_config['alibi'] and self.attn_config['attn_impl'] not in [
'torch', 'triton'
]:
raise NotImplementedError(
'alibi only implemented with torch and triton attention.')
if self.attn_config['attn_uses_sequence_id'] and self.attn_config[
'attn_impl'] not in ['torch', 'triton']:
raise NotImplementedError(
'attn_uses_sequence_id only implemented with torch '
'and triton attention.')
if self.embedding_fraction > 1 or self.embedding_fraction <= 0:
raise ValueError(
'model.embedding_fraction must be between 0 (exclusive) '
'and 1 (inclusive)!')
if isinstance(self.logit_scale,
str) and self.logit_scale != 'inv_sqrt_d_model':
raise ValueError(
f"self.logit_scale={self.logit_scale!r} is not recognized as "
"an option; use numeric value or 'inv_sqrt_d_model'.")
if self.init_config.get('name', None) is None:
raise ValueError(
f"self.init_config={self.init_config!r} 'name' needs to be set."
)
if not self.learned_pos_emb and (not self.attn_config['alibi']):
warnings.warn(
'Positional information not being provided to the model.',
stacklevel=2)
if self.fc_type == 'te' or self.ffn_config['ffn_type'] == 'te_ln_mlp':
try:
# pylint: disable=import-outside-toplevel
import transformer_engine.pytorch as te
del te
except Exception as exc:
raise ImportError(
'TransformerEngine import fail. `fc_type: te` requires '
'TransformerEngine be installed. '
'The required version of transformer_engine also requires '
'FlashAttention v1.0.6 is installed:\n'
'pip install flash-attn==1.0.6 --no-build-isolation \n'
'pip install git+https://github.com/NVIDIA/TransformerEngine.git@144e4888b2cdd60bd52e706d5b7a79cb9c1a7156'
) from exc
if self.ffn_config['ffn_type'] == 'mptmlp':
self.ffn_config['fc_type'] = self.fc_type
elif self.ffn_config['ffn_type'] == 'te_ln_mlp':
self.ffn_config['bias'] = not self.no_bias