huangyanpei/VLA
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update Wall-X to 1.1.0 (#104)

Browse Source
This commit is contained in:
Starrick Liu
2026-06-15 11:40:00 +08:00
committed by GitHub
parent e23a586846
commit 72834e7de5
200 changed files with 33916 additions and 16771 deletions
Download Patch File Download Diff File Expand all files Collapse all files
wall_x/model/qact/qwen2_5/__init__.py
+2
View File
@@ -0,0 +1,2 @@
from .configuration_qwen2_5_vl import Qwen2_5_VLConfig
from .modeling_qwen2_5_vl_act import Qwen2_5_VLMoEForAction, Qwen2_5_VLMoEModel
wall_x/model/qact/qwen2_5/adapter.py
+56
View File
@@ -0,0 +1,56 @@
"""Qwen2.5 VLA adapter - variant-specific overrides on top of VLAdapter."""
from wall_x.model.registry import register_model
from wall_x.trainer.adapters.vla_model_adapter import VLAdapter
@register_model("qwen2_5")
class Qwen2_5Adapter(VLAdapter):
MODEL_TYPE = "qwen2_5"
@classmethod
def model_class(cls):
from wall_x.model.qact.qwen2_5 import Qwen2_5_VLMoEForAction
return Qwen2_5_VLMoEForAction
@classmethod
def config_class(cls):
from wall_x.model.qact.qwen2_5 import Qwen2_5_VLConfig
return Qwen2_5_VLConfig
@classmethod
def inference_model_class(cls):
return cls.model_class()
def get_transformer_layer_cls(self):
layer_classes = set()
try:
from transformers.models.qwen2_vl.modeling_qwen2_vl import (
Qwen2VLDecoderLayer,
)
layer_classes.add(Qwen2VLDecoderLayer)
except ImportError:
pass
try:
from wall_x.model.qact.qwen2_5.modeling_qwen2_5_vl import (
Qwen2_5_VLDecoderLayer,
)
layer_classes.add(Qwen2_5_VLDecoderLayer)
except ImportError:
pass
return layer_classes if layer_classes else None
@staticmethod
def log_attention_implementation(logger, model):
logger.info(
f"*** model attention implementation: "
f"{model.model._attn_implementation} ***"
)
logger.info(
f"*** model.visual attention implementation: "
f"{model.visual.config._attn_implementation} ***"
)
wall_x/model/qact/qwen2_5/configuration_qwen2_5_vl.py
+357
View File
@@ -0,0 +1,357 @@
# !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# This file was automatically generated from src/transformers/models/qwen2_5_vl/modular_qwen2_5_vl.py.
# Do NOT edit this file manually as any edits will be overwritten by the generation of
# the file from the modular. If any change should be done, please apply the change to the
# modular_qwen2_5_vl.py file directly. One of our CI enforces this.
# !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# coding=utf-8
# Copyright 2025 The Qwen Team and The HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
from transformers.configuration_utils import PretrainedConfig
from transformers.modeling_rope_utils import rope_config_validation
logger = logging.getLogger(__name__)
class Qwen2_5_VLVisionConfig(PretrainedConfig):
model_type = "qwen2_5_vl"
base_config_key = "vision_config"
def __init__(
self,
depth=32,
hidden_size=3584,
hidden_act="silu",
intermediate_size=3420,
num_heads=16,
in_channels=3,
patch_size=14,
spatial_merge_size=2,
temporal_patch_size=2,
tokens_per_second=4,
window_size=112,
out_hidden_size=3584,
fullatt_block_indexes=[7, 15, 23, 31],
initializer_range=0.02,
_attn_implementation="flash_attention_2",
attn_deterministic=False,
**kwargs,
):
super().__init__(**kwargs)
self.depth = depth
self.hidden_size = hidden_size
self.hidden_act = hidden_act
self.intermediate_size = intermediate_size
self.num_heads = num_heads
self.in_channels = in_channels
self.patch_size = patch_size
self.spatial_merge_size = spatial_merge_size
self.temporal_patch_size = temporal_patch_size
self.tokens_per_second = tokens_per_second
self.window_size = window_size
self.fullatt_block_indexes = fullatt_block_indexes
self.out_hidden_size = out_hidden_size
self.initializer_range = initializer_range
self._attn_implementation = _attn_implementation
self.attn_deterministic = attn_deterministic
class Qwen2_5_VLConfig(PretrainedConfig):
r"""
This is the configuration class to store the configuration of a [`Qwen2_5_VLModel`]. It is used to instantiate a
Qwen2-VL model according to the specified arguments, defining the model architecture. Instantiating a configuration
with the defaults will yield a similar configuration to that of
Qwen2-VL-7B-Instruct [Qwen/Qwen2-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct).
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 152064):
Vocabulary size of the Qwen2_5_VL model. Defines the number of different tokens that can be represented by the
`inputs_ids` passed when calling [`Qwen2_5_VLModel`]
hidden_size (`int`, *optional*, defaults to 8192):
Dimension of the hidden representations.
intermediate_size (`int`, *optional*, defaults to 29568):
Dimension of the MLP representations.
num_hidden_layers (`int`, *optional*, defaults to 80):
Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 64):
Number of attention heads for each attention layer in the Transformer encoder.
num_key_value_heads (`int`, *optional*, defaults to 8):
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 `32`.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) in the decoder.
max_position_embeddings (`int`, *optional*, defaults to 32768):
The maximum sequence length that this model might ever be used with.
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
rms_norm_eps (`float`, *optional*, defaults to 1e-05):
The epsilon used by the rms normalization layers.
use_cache (`bool`, *optional*, defaults to `True`):
Whether or not the model should return the last key/values attentions (not used by all models). Only
relevant if `config.is_decoder=True`.
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
Whether the model's input and output word embeddings should be tied.
rope_theta (`float`, *optional*, defaults to 1000000.0):
The base period of the RoPE embeddings.
use_sliding_window (`bool`, *optional*, defaults to `False`):
Whether to use sliding window attention.
sliding_window (`int`, *optional*, defaults to 4096):
Sliding window attention (SWA) window size. If not specified, will default to `4096`.
max_window_layers (`int`, *optional*, defaults to 80):
The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
attention_dropout (`float`, *optional*, defaults to 0.0):
The dropout ratio for the attention probabilities.
vision_config (`Dict`, *optional*):
The config for the visual encoder initialization.
rope_scaling (`Dict`, *optional*):
Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
accordingly.
Expected contents:
`rope_type` (`str`):
The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
'llama3'], with 'default' being the original RoPE implementation.
`factor` (`float`, *optional*):
Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
most scaling types, a `factor` of x will enable the model to handle sequences of length x *
original maximum pre-trained length.
`original_max_position_embeddings` (`int`, *optional*):
Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
pretraining.
`attention_factor` (`float`, *optional*):
Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
computation. If unspecified, it defaults to value recommended by the implementation, using the
`factor` field to infer the suggested value.
`beta_fast` (`float`, *optional*):
Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
ramp function. If unspecified, it defaults to 32.
`beta_slow` (`float`, *optional*):
Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
ramp function. If unspecified, it defaults to 1.
`short_factor` (`List[float]`, *optional*):
Only used with 'longrope'. The scaling factor to be applied to short contexts (<
`original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
size divided by the number of attention heads divided by 2
`long_factor` (`List[float]`, *optional*):
Only used with 'longrope'. The scaling factor to be applied to long contexts (<
`original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
size divided by the number of attention heads divided by 2
`low_freq_factor` (`float`, *optional*):
Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
`high_freq_factor` (`float`, *optional*):
Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
```python
>>> from transformers import Qwen2_5_VLForConditionalGeneration, Qwen2_5_VLConfig
>>> # Initializing a Qwen2_5_VL style configuration
>>> configuration = Qwen2_5_VLConfig()
>>> # Initializing a model from the Qwen2-VL-7B style configuration
>>> model = Qwen2_5_VLForConditionalGeneration(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```"""
model_type = "qwen2_5_vl"
sub_configs = {"vision_config": Qwen2_5_VLVisionConfig}
keys_to_ignore_at_inference = ["past_key_values"]
# Default tensor parallel plan for base model `Qwen2_5_VL`
base_model_tp_plan = {
"layers.*.self_attn.q_proj": "colwise",
"layers.*.self_attn.k_proj": "colwise",
"layers.*.self_attn.v_proj": "colwise",
"layers.*.self_attn.o_proj": "rowwise",
"layers.*.mlp.gate_proj": "colwise",
"layers.*.mlp.up_proj": "colwise",
"layers.*.mlp.down_proj": "rowwise",
}
base_model_pp_plan = {
"embed_tokens": (["input_ids"], ["inputs_embeds"]),
"layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
"norm": (["hidden_states"], ["hidden_states"]),
}
def __init__(
self,
vocab_size=152064,
hidden_size=8192,
action_hidden_size=2048,
state_hidden_size=2048,
intermediate_size=29568,
num_hidden_layers=80,
num_attention_heads=64,
num_key_value_heads=8,
hidden_act="silu",
max_position_embeddings=32768,
initializer_range=0.02,
rms_norm_eps=1e-05,
use_cache=True,
tie_word_embeddings=False,
rope_theta=1000000.0,
use_sliding_window=False,
sliding_window=4096,
max_window_layers=80,
attention_dropout=0.0,
vision_config=None,
rope_scaling=None,
num_experts=4,
experts=None,
dof_config=None,
noise_scheduler=None,
dim_inputs=(1536, 1536),
attention_moe=False,
mlp_moe=False,
norm_moe=False,
mot_opt=False,
ar_loss_weight=1.0,
use_state_string_representation=False,
use_adarms=False,
proj_with_mask=True,
adarms_cond_dim=None,
action_horizon_flow=32,
causal_action_attention_mask=False,
use_flow_action_expert=True,
use_x_pred=False,
attn_deterministic=False,
use_x_loss=False,
**kwargs,
):
# Compatibility with newer transformers versions (5.x):
# - Older versions: super() sets self.pad_token_id; override it with the saved value so kwargs are preserved
# - Newer versions: super() does not set self.pad_token_id; assign it afterward
_pad_token_id = kwargs.pop("pad_token_id", None)
self.vocab_size = vocab_size
self.max_position_embeddings = max_position_embeddings
self.hidden_size = hidden_size
self.action_hidden_size = action_hidden_size
self.state_hidden_size = state_hidden_size
self.intermediate_size = intermediate_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.use_sliding_window = use_sliding_window
self.sliding_window = sliding_window
self.max_window_layers = max_window_layers
# for backward compatibility
if num_key_value_heads is None:
num_key_value_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.hidden_act = hidden_act
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.use_cache = use_cache
self.rope_theta = rope_theta
self.attention_dropout = attention_dropout
self.rope_scaling = rope_scaling
self.num_experts = num_experts
self.experts = experts
self.dof_config = dof_config
self.noise_scheduler = noise_scheduler
self.dim_inputs = tuple(dim_inputs)
self.attention_moe = attention_moe
self.mlp_moe = mlp_moe
self.norm_moe = norm_moe
self.mot_opt = mot_opt
self.ar_loss_weight = ar_loss_weight
self.use_state_string_representation = use_state_string_representation
self.use_adarms = use_adarms
self.adarms_cond_dim = adarms_cond_dim
self.proj_with_mask = proj_with_mask
self.use_flow_action_expert = use_flow_action_expert
self.action_horizon_flow = action_horizon_flow
self.causal_action_attention_mask = causal_action_attention_mask
self.use_flow_action_expert = use_flow_action_expert
self.use_x_pred = use_x_pred
self.attn_deterministic = attn_deterministic
self.use_x_loss = use_x_loss
# Validate the correctness of rotary position embeddings parameters
# BC: if there is a 'type' field, move it to 'rope_type'.
# and change type from 'mrope' to 'default' because `mrope` does defeault RoPE calculations
# one can set it to "linear"/"dynamic" etc. to have scaled RoPE
# TODO: @raushan update config in the hub
if self.rope_scaling is not None and "type" in self.rope_scaling:
if self.rope_scaling["type"] == "mrope":
self.rope_scaling["type"] = "default"
self.rope_scaling["rope_type"] = self.rope_scaling["type"]
rope_config_validation(self, ignore_keys={"mrope_section"})
super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)
# Assign pad_token_id as described above, overriding older super() values or filling newer missing attributes
self.pad_token_id = _pad_token_id
# move vision config initialization after super init to avoid recursively set in latest transformers version
# TODO: make it better
if isinstance(vision_config, dict):
self.vision_config = self.sub_configs["vision_config"](**vision_config)
elif vision_config is None:
self.vision_config = self.sub_configs["vision_config"]()
def update_model_config(self, train_config):
"""Update model configuration from training config.
This method updates the model configuration with training-specific
settings such as action horizon, DOF config, attention implementation, etc.
Args:
train_config: dict containing training configuration parameters.
"""
self.use_state_string_representation = train_config["data"].get(
"use_state_string_representation", False
)
self.ar_loss_weight = train_config.get("ar_loss_weight", 1.0)
self.dof_config = train_config["dof_config"]
self.agent_pos_config = train_config["agent_pos_config"]
self.action_horizon_flow = train_config["data"].get("action_horizon_flow", 32)
if train_config.get("_attn_implementation", None) is not None:
self._attn_implementation = train_config["_attn_implementation"]
if train_config.get("attn_deterministic", None) is not None:
self.attn_deterministic = train_config["attn_deterministic"]
self.vision_config.attn_deterministic = train_config["attn_deterministic"]
logger.debug("Attention is using deterministic kernel for this run")
else:
self.attn_deterministic = True
self.vision_config.attn_deterministic = True
if train_config.get("noise_scheduler", None) is not None:
self.noise_scheduler = train_config["noise_scheduler"]
__all__ = ["Qwen2_5_VLConfig"]
wall_x/model/qact/qwen2_5/inference_mixin.py
+1039
View File
File diff suppressed because it is too large Load Diff
wall_x/model/qact/qwen2_5/modeling_qwen2_5_vl.py
+2900
View File
File diff suppressed because it is too large Load Diff
wall_x/model/qact/qwen2_5/modeling_qwen2_5_vl_act.py
+3065
View File
File diff suppressed because it is too large Load Diff