apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 00:33:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix and refine vl (#2866)

Browse Source 
* refine vl config

* delete attn_sep

* fix vl accuracy
This commit is contained in:
Yuanle Liu
2025-07-16 20:59:28 +08:00
committed by GitHub
parent aa76085d1f
commit 63d6e7ce06
11 changed files with 63 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
fastdeploy/config.py
View File

@@ -21,7 +21,6 @@ from enum import Enum
from typing import Literal, Optional from typing import Literal, Optional
from paddleformers.transformers.configuration_utils import PretrainedConfig from paddleformers.transformers.configuration_utils import PretrainedConfig
from paddleformers.trl import llm_utils
from fastdeploy import envs from fastdeploy import envs
from fastdeploy.model_executor.layers.quantization.quant_base import \ from fastdeploy.model_executor.layers.quantization.quant_base import \
@@ -52,10 +51,6 @@ PRETRAINED_INIT_CONFIGURATION = {
"initializer_range" : 0.02, "initializer_range" : 0.02,
"max_position_embeddings" : 512, "max_position_embeddings" : 512,
"quantization_config" : None, "quantization_config" : None,
"use_recompute_resampler":False,
"use_temporal_conv":True,
"resampler_fuse_rms_norm":False,
"freq_allocation":20,
"tie_word_embeddings" : False, "tie_word_embeddings" : False,
"rms_norm_eps" : 1e-5, "rms_norm_eps" : 1e-5,
"moe_num_experts" : None, "moe_num_experts" : None,
@@ -84,9 +79,6 @@ class ModelConfig:
self.min_length = 1 self.min_length = 1
self.model_name_or_path = "" self.model_name_or_path = ""
self.im_patch_id = (
100295 # multimodality, TODO(liuyuanle): read from config.json
)
self.is_quantized = False self.is_quantized = False
self.max_model_len = 0 self.max_model_len = 0
self.dtype = "" self.dtype = ""
@@ -130,10 +122,9 @@ class ParallelConfig:
self.moe_phase = MoEPhase.PREFILL # Generation phase self.moe_phase = MoEPhase.PREFILL # Generation phase
self.msg_queue_id = 1 # mesage queue id self.msg_queue_id = 1 # mesage queue id
tensor_parallel_rank, tensor_parallel_size = llm_utils.init_dist_env() self.tensor_parallel_rank = 0 # TP rank ID
self.tensor_parallel_rank = tensor_parallel_rank # TP rank ID self.tensor_parallel_size = 1 # TP degree
self.tensor_parallel_size = tensor_parallel_size # TP degree self.expert_parallel_rank = 0 # EP rank ID
self.expert_parallel_rank = int(tensor_parallel_rank / tensor_parallel_size) # EP rank ID
self.expert_parallel_size = 1 # EP degree self.expert_parallel_size = 1 # EP degree
# The embedding weight distributed on your gpu cards is divided by row or column. # The embedding weight distributed on your gpu cards is divided by row or column.
# Defaults to False means divide by row. When vocab_size can not be divided by world_size # Defaults to False means divide by row. When vocab_size can not be divided by world_size

1
fastdeploy/model_executor/layers/embeddings.py
View File

@@ -62,7 +62,6 @@ class VocabParallelEmbedding(nn.Layer):
self.use_ep: bool = fd_config.parallel_config.use_ep self.use_ep: bool = fd_config.parallel_config.use_ep
self.hidden_dropout_prob: float = fd_config.model_config.hidden_dropout_prob self.hidden_dropout_prob: float = fd_config.model_config.hidden_dropout_prob
self.initializer_range: float = fd_config.model_config.initializer_range self.initializer_range: float = fd_config.model_config.initializer_range
self.sequence_parallel: bool = fd_config.parallel_config.sequence_parallel
self.max_position_embeddings: int = fd_config.model_config.max_position_embeddings self.max_position_embeddings: int = fd_config.model_config.max_position_embeddings
self.tie_word_embeddings: bool = fd_config.model_config.tie_word_embeddings self.tie_word_embeddings: bool = fd_config.model_config.tie_word_embeddings
self.params_dtype: str = params_dtype self.params_dtype: str = params_dtype

16
fastdeploy/model_executor/layers/rotary_embedding.py
View File

@@ -423,11 +423,11 @@ class ErnieVlRotaryEmbedding3D:
def get_rope_3d( def get_rope_3d(
rotary_dim: int, rotary_dim: int,
base: 10000, base: float,
position_ids, position_ids: paddle.Tensor,
paritial_rotary_factor: 1, partial_rotary_factor: float,
max_position: 131072, max_position: int,
freq_allocation: 2, freq_allocation: int,
) -> paddle.Tensor: ) -> paddle.Tensor:
""" """
Pre-calculate rotary position embedding for position_ids. Pre-calculate rotary position embedding for position_ids.
@@ -435,19 +435,19 @@ def get_rope_3d(
Args: Args:
rotary_dim (int): rotary_dim (int):
Dimension of rotary embeddings (head dimension) Dimension of rotary embeddings (head dimension)
base (float, optional): base (float):
Base value used to compute the inverse frequencies. Base value used to compute the inverse frequencies.
Default: 10000.0. Default: 10000.0.
position_ids (paddle.Tensor): position_ids (paddle.Tensor):
Tensor containing position indices of input tokens. Tensor containing position indices of input tokens.
partial_rotary_factor (int, optional): partial_rotary_factor (float):
Factor controlling partial rotary application. Factor controlling partial rotary application.
Default: 1 (apply to all dimensions). Default: 1 (apply to all dimensions).
max_position: Maximum position index to precompute. max_position: Maximum position index to precompute.
freq_allocation: Number of rotary dimensions allocated to temporal axis freq_allocation: Number of rotary dimensions allocated to temporal axis
""" """
rotary_emb3d_layer = ErnieVlRotaryEmbedding3D(rotary_dim, base, rotary_emb3d_layer = ErnieVlRotaryEmbedding3D(rotary_dim, base,
paritial_rotary_factor, partial_rotary_factor,
max_position, max_position,
freq_allocation) freq_allocation)
rotary_emb_3d = rotary_emb3d_layer(position_ids) rotary_emb_3d = rotary_emb3d_layer(position_ids)

1
fastdeploy/model_executor/models/ernie4_5_moe.py
View File

@@ -71,6 +71,7 @@ class Ernie4_5_MLP(nn.Layer):
input_size=intermediate_size, input_size=intermediate_size,
output_size=fd_config.model_config.hidden_size, output_size=fd_config.model_config.hidden_size,
with_bias=False, with_bias=False,
reduce_results=reduce_results,
) )
self.act_fn = SiluAndMul( self.act_fn = SiluAndMul(

2
fastdeploy/model_executor/models/ernie4_5_vl/dfnrope/configuration.py
View File

@@ -46,7 +46,6 @@ class DFNRopeVisionTransformerConfig(PretrainedConfig):
attn_implementation="eager", # new added attn_implementation="eager", # new added
pp_data_balance=False, pp_data_balance=False,
recompute=False, recompute=False,
attn_sep=False,
vit_first_fwd_bsz=128, vit_first_fwd_bsz=128,
vit_num_recompute_layers=10000, vit_num_recompute_layers=10000,
**kwargs, **kwargs,
@@ -65,6 +64,5 @@ class DFNRopeVisionTransformerConfig(PretrainedConfig):
self.attn_implementation = attn_implementation self.attn_implementation = attn_implementation
self.pp_data_balance = pp_data_balance self.pp_data_balance = pp_data_balance
self.recompute = recompute self.recompute = recompute
self.attn_sep = attn_sep
self.vit_first_fwd_bsz = vit_first_fwd_bsz self.vit_first_fwd_bsz = vit_first_fwd_bsz
self.vit_num_recompute_layers = vit_num_recompute_layers self.vit_num_recompute_layers = vit_num_recompute_layers

69
fastdeploy/model_executor/models/ernie4_5_vl/dfnrope/modeling.py
View File

@@ -143,30 +143,6 @@ def apply_rotary_pos_emb_vision(tensor: paddle.Tensor,
return output return output
def qkv_reshard_head(tensor, group):
"""
将qkv在seq维度拼接后一起做切分维度的转换
"""
parallelism = group.nranks
qkv_seqlen, head_num, head_dim = tensor.shape
tensor = tensor.transpose(perm=[1, 0, 2]).contiguous()
out = _AllToAll.apply(tensor, group)
out = paddle.split(out, parallelism, axis=0)
output_q = []
output_k = []
output_v = []
for output_i in out:
outout = output_i.transpose(perm=[1, 0, 2]).contiguous()
output = paddle.split(outout, 3, axis=0)
output_q.append(output[0])
output_k.append(output[1])
output_v.append(output[2])
q = paddle.concat(output_q, axis=0)
k = paddle.concat(output_k, axis=0)
v = paddle.concat(output_v, axis=0)
return q, k, v
class VisionFlashAttention2(nn.Layer): class VisionFlashAttention2(nn.Layer):
"""_summary_ """_summary_
@@ -211,7 +187,6 @@ class VisionFlashAttention2(nn.Layer):
hidden_states: paddle.Tensor, hidden_states: paddle.Tensor,
cu_seqlens: paddle.Tensor, cu_seqlens: paddle.Tensor,
rotary_pos_emb: paddle.Tensor = None, rotary_pos_emb: paddle.Tensor = None,
attn_sep=False,
) -> paddle.Tensor: ) -> paddle.Tensor:
"""_summary_ """_summary_
@@ -229,13 +204,6 @@ class VisionFlashAttention2(nn.Layer):
-1]).transpose(perm=[1, 0, 2, 3]) -1]).transpose(perm=[1, 0, 2, 3])
q, k, v = qkv.unbind(axis=0) q, k, v = qkv.unbind(axis=0)
if attn_sep:
hcg = get_hcg()
mp_group = hcg.get_model_parallel_group()
qkv = paddle.concat([q, k, v], axis=0)
q, k, v = qkv_reshard_head(qkv, mp_group)
seq_length = q.shape[0]
q = apply_rotary_pos_emb_vision(q.unsqueeze(axis=0), q = apply_rotary_pos_emb_vision(q.unsqueeze(axis=0),
rotary_pos_emb).squeeze(axis=0) rotary_pos_emb).squeeze(axis=0)
k = apply_rotary_pos_emb_vision(k.unsqueeze(axis=0), k = apply_rotary_pos_emb_vision(k.unsqueeze(axis=0),
@@ -256,10 +224,7 @@ class VisionFlashAttention2(nn.Layer):
max_seqlen, max_seqlen,
scale=softmax_scale, # TODO: 需要手动加上 scale=softmax_scale, # TODO: 需要手动加上
)[0].squeeze(0).reshape([seq_length, -1])) )[0].squeeze(0).reshape([seq_length, -1]))
if attn_sep:
out = _AllToAll.apply(attn_output, mp_group)
out = paddle.split(out, mp_group.nranks, axis=0)
attn_output = paddle.concat(out, axis=1)
attn_output = attn_output.astype(paddle.float32) attn_output = attn_output.astype(paddle.float32)
attn_output = self.proj(attn_output) attn_output = self.proj(attn_output)
return attn_output return attn_output
@@ -389,7 +354,7 @@ class DFNRopeVisionBlock(nn.Layer):
nn (_type_): _description_ nn (_type_): _description_
""" """
def __init__(self, config, attn_implementation: str = "sdpa") -> None: def __init__(self, config, tensor_parallel_degree: int, attn_implementation: str = "sdpa") -> None:
"""_summary_ """_summary_
Args: Args:
@@ -404,19 +369,18 @@ class DFNRopeVisionBlock(nn.Layer):
self.attn = VisionFlashAttention2( self.attn = VisionFlashAttention2(
config.embed_dim, config.embed_dim,
num_heads=config.num_heads, num_heads=config.num_heads,
tensor_parallel_degree=config.tensor_parallel_degree) tensor_parallel_degree=tensor_parallel_degree)
self.mlp = VisionMlp( self.mlp = VisionMlp(
dim=config.embed_dim, dim=config.embed_dim,
hidden_dim=mlp_hidden_dim, hidden_dim=mlp_hidden_dim,
hidden_act=config.hidden_act, hidden_act=config.hidden_act,
tensor_parallel_degree=config.tensor_parallel_degree) tensor_parallel_degree=tensor_parallel_degree)
self.config = config self.config = config
def forward(self, def forward(self,
hidden_states, hidden_states,
cu_seqlens, cu_seqlens,
rotary_pos_emb, rotary_pos_emb) -> paddle.Tensor:
attn_sep=False) -> paddle.Tensor:
"""_summary_ """_summary_
Args: Args:
@@ -431,7 +395,6 @@ class DFNRopeVisionBlock(nn.Layer):
self.norm1(hidden_states), self.norm1(hidden_states),
cu_seqlens=cu_seqlens, cu_seqlens=cu_seqlens,
rotary_pos_emb=rotary_pos_emb, rotary_pos_emb=rotary_pos_emb,
attn_sep=attn_sep,
) )
hidden_states = hidden_states + self.mlp(self.norm2(hidden_states)) hidden_states = hidden_states + self.mlp(self.norm2(hidden_states))
return hidden_states return hidden_states
@@ -490,26 +453,26 @@ class DFNRopeVisionTransformerPretrainedModel(PretrainedModel):
config_class = DFNRopeVisionTransformerConfig config_class = DFNRopeVisionTransformerConfig
def __init__(self, config, prefix_name: str = "") -> None: def __init__(self, config, prefix_name: str = "") -> None:
super().__init__(config) super().__init__(config.vision_config)
self.spatial_merge_size = config.spatial_merge_size self.spatial_merge_size = config.vision_config.spatial_merge_size
self.prefix_name = prefix_name self.prefix_name = prefix_name
self.patch_embed = PatchEmbed( self.patch_embed = PatchEmbed(
patch_size=config.patch_size, patch_size=config.vision_config.patch_size,
in_channels=config.in_channels, in_channels=config.vision_config.in_channels,
embed_dim=config.embed_dim, embed_dim=config.vision_config.embed_dim,
) )
head_dim = config.embed_dim // config.num_heads head_dim = config.vision_config.embed_dim // config.vision_config.num_heads
self.rotary_pos_emb = VisionRotaryEmbedding(head_dim // 2) self.rotary_pos_emb = VisionRotaryEmbedding(head_dim // 2)
self.blocks = nn.LayerList( self.blocks = nn.LayerList(
[DFNRopeVisionBlock(config) for _ in range(config.depth)]) [DFNRopeVisionBlock(config.vision_config, config.pretrained_config.tensor_parallel_degree) for _ in range(config.vision_config.depth)])
assert ( assert (
config.hidden_size == config.embed_dim config.vision_config.hidden_size == config.vision_config.embed_dim
), "in DFNRope, vit's config.hidden must be equal to config.embed_dim" ), "in DFNRope, vit's config.hidden must be equal to config.embed_dim"
# self.merger = PatchMerger(dim=config.hidden_size, context_dim=config.embed_dim) # self.merger = PatchMerger(dim=config.hidden_size, context_dim=config.embed_dim)
self.ln = nn.LayerNorm(config.hidden_size, epsilon=1e-6) self.ln = nn.LayerNorm(config.vision_config.hidden_size, epsilon=1e-6)
def get_dtype(self) -> paddle.dtype: def get_dtype(self) -> paddle.dtype:
"""_summary_ """_summary_
@@ -593,7 +556,6 @@ class DFNRopeVisionTransformerPretrainedModel(PretrainedModel):
else: else:
cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0) cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
attn_sep = getattr(self.config, "attn_sep", False)
vit_num_recompute_layers = getattr(self.config, vit_num_recompute_layers = getattr(self.config,
"vit_num_recompute_layers", "vit_num_recompute_layers",
self.config.depth) self.config.depth)
@@ -601,13 +563,12 @@ class DFNRopeVisionTransformerPretrainedModel(PretrainedModel):
for idx, blk in enumerate(self.blocks): for idx, blk in enumerate(self.blocks):
if self.config.recompute and self.training and idx < vit_num_recompute_layers: if self.config.recompute and self.training and idx < vit_num_recompute_layers:
hidden_states = recompute(blk, hidden_states, cu_seqlens, hidden_states = recompute(blk, hidden_states, cu_seqlens,
rotary_pos_emb, attn_sep) rotary_pos_emb)
else: else:
hidden_states = blk( hidden_states = blk(
hidden_states, hidden_states,
cu_seqlens=cu_seqlens, cu_seqlens=cu_seqlens,
rotary_pos_emb=rotary_pos_emb, rotary_pos_emb=rotary_pos_emb,
attn_sep=attn_sep,
) )
# ret = self.merger(hidden_states) # ret = self.merger(hidden_states)

11
fastdeploy/model_executor/models/ernie4_5_vl/ernie4_5_vl_moe.py
View File

@@ -123,6 +123,7 @@ class Ernie4_5_VLMoE(nn.Layer):
fd_config=fd_config, fd_config=fd_config,
intermediate_size=fd_config.model_config.intermediate_size, intermediate_size=fd_config.model_config.intermediate_size,
prefix=f"{prefix}", prefix=f"{prefix}",
reduce_results=False,
) )
assert image_moe_layer_start_index <= image_moe_layer_end_index assert image_moe_layer_start_index <= image_moe_layer_end_index
@@ -155,6 +156,7 @@ class Ernie4_5_VLMoE(nn.Layer):
fd_config=fd_config, fd_config=fd_config,
intermediate_size=fd_config.model_config.intermediate_size, intermediate_size=fd_config.model_config.intermediate_size,
prefix=f"{prefix}", prefix=f"{prefix}",
reduce_results=False,
) )
self.num_shared_experts = fd_config.model_config.moe_num_shared_experts self.num_shared_experts = fd_config.model_config.moe_num_shared_experts
@@ -471,8 +473,7 @@ class Ernie4_5_VLMoeForConditionalGeneration(ModelForCasualLM):
""" """
super(Ernie4_5_VLMoeForConditionalGeneration, self).__init__(fd_config) super(Ernie4_5_VLMoeForConditionalGeneration, self).__init__(fd_config)
# ----------- vision model ------------ # ----------- vision model ------------
vision_config = fd_config.model_config.vision_config self.vision_model = self._init_vision_model(fd_config.model_config)
self.vision_model = self._init_vision_model(vision_config)
# ----------- resampler_model ------------ # ----------- resampler_model ------------
self.resampler_model = self._init_resampler_model_model( self.resampler_model = self._init_resampler_model_model(
fd_config.model_config fd_config.model_config
@@ -490,12 +491,12 @@ class Ernie4_5_VLMoeForConditionalGeneration(ModelForCasualLM):
) )
self.tie_word_embeddings = fd_config.model_config.tie_word_embeddings self.tie_word_embeddings = fd_config.model_config.tie_word_embeddings
def _init_vision_model(self, vision_config) -> nn.Layer: def _init_vision_model(self, model_config) -> nn.Layer:
from fastdeploy.model_executor.models.ernie4_5_vl.dfnrope.modeling import \ from fastdeploy.model_executor.models.ernie4_5_vl.dfnrope.modeling import \
DFNRopeVisionTransformerPretrainedModel DFNRopeVisionTransformerPretrainedModel
vision_model = DFNRopeVisionTransformerPretrainedModel( vision_model = DFNRopeVisionTransformerPretrainedModel(
vision_config, prefix_name="vision_model" model_config, prefix_name="vision_model"
) )
vision_model = paddle.amp.decorate( vision_model = paddle.amp.decorate(
models=vision_model, level="O2", dtype="bfloat16" models=vision_model, level="O2", dtype="bfloat16"
@@ -508,7 +509,7 @@ class Ernie4_5_VLMoeForConditionalGeneration(ModelForCasualLM):
VariableResolutionResamplerModel VariableResolutionResamplerModel
resampler_model = VariableResolutionResamplerModel( resampler_model = VariableResolutionResamplerModel(
model_config.pixel_hidden_size, model_config.vision_config.hidden_size,
model_config.hidden_size, model_config.hidden_size,
model_config.spatial_conv_size, model_config.spatial_conv_size,
model_config.temporal_conv_size, model_config.temporal_conv_size,

15
fastdeploy/model_executor/models/ernie4_5_vl/modeling_resampler.py
View File

@@ -104,7 +104,7 @@ class RMSNorm(nn.Layer):
self.variance_epsilon = config.rms_norm_eps self.variance_epsilon = config.rms_norm_eps
self.config = config self.config = config
if config.sequence_parallel: if getattr(config, "sequence_parallel", False):
mark_as_sequence_parallel_parameter(self.weight) mark_as_sequence_parallel_parameter(self.weight)
def forward(self, hidden_states): def forward(self, hidden_states):
@@ -118,7 +118,6 @@ class RMSNorm(nn.Layer):
Tensor: Normalized output tensor of same shape as input Tensor: Normalized output tensor of same shape as input
Note: Note:
- Uses fused kernel if config.fuse_rms_norm is True for better performance
- Otherwise computes RMSNorm manually: - Otherwise computes RMSNorm manually:
1. Compute variance of features 1. Compute variance of features
2. Apply reciprocal square root normalization 2. Apply reciprocal square root normalization
@@ -146,9 +145,9 @@ class VariableResolutionResamplerModel(nn.Layer):
self.config = config self.config = config
self.spatial_conv_size = spatial_conv_size self.spatial_conv_size = spatial_conv_size
self.temporal_conv_size = temporal_conv_size self.temporal_conv_size = temporal_conv_size
self.use_recompute_resampler = config.use_recompute_resampler self.use_recompute_resampler = False
self.use_temporal_conv = config.use_temporal_conv self.use_temporal_conv = True
self.tensor_parallel_degree = config.tensor_parallel_degree self.tensor_parallel_degree = config.pretrained_config.tensor_parallel_degree
self.prefix_name = prefix_name self.prefix_name = prefix_name
# for 空间四合一 # for 空间四合一
@@ -165,7 +164,7 @@ class VariableResolutionResamplerModel(nn.Layer):
input_is_parallel=True, input_is_parallel=True,
has_bias=True, has_bias=True,
fuse_matmul_bias=True, fuse_matmul_bias=True,
) if config.tensor_parallel_degree > 1 else nn.Linear( ) if self.tensor_parallel_degree > 1 else nn.Linear(
self.spatial_dim, self.spatial_dim)), self.spatial_dim, self.spatial_dim)),
nn.GELU(), nn.GELU(),
nn.Linear(self.spatial_dim, self.spatial_dim), nn.Linear(self.spatial_dim, self.spatial_dim),
@@ -184,11 +183,9 @@ class VariableResolutionResamplerModel(nn.Layer):
out_config = deepcopy(config) out_config = deepcopy(config)
out_config.hidden_size = out_dim out_config.hidden_size = out_dim
# Note(GuoxiaWang): fuse can reduce gpu peak memory
out_config.fuse_rms_norm = out_config.resampler_fuse_rms_norm
self.after_norm = RMSNorm(out_config) self.after_norm = RMSNorm(out_config)
if config.tensor_parallel_degree > 1: if self.tensor_parallel_degree > 1:
for idx in [2, 3]: for idx in [2, 3]:
mark_as_sequence_parallel_parameter( mark_as_sequence_parallel_parameter(
self.spatial_linear[idx].weight) self.spatial_linear[idx].weight)

2
fastdeploy/rl/rollout_config.py
View File

@@ -102,4 +102,4 @@ class RolloutModelConfig:
def initialize(self): def initialize(self):
"""Initialize the final fd config""" """Initialize the final fd config"""
return initialize_fd_config(self, self.tensor_parallel_size, 0) return initialize_fd_config(self, ranks=self.tensor_parallel_size, local_rank=0)

13
fastdeploy/worker/gpu_model_runner.py
View File

@@ -1447,20 +1447,15 @@ class GPUModelRunner(ModelRunnerBase):
self.fd_config.model_config.tensor_parallel_degree = self.parallel_config.tensor_parallel_size self.fd_config.model_config.tensor_parallel_degree = self.parallel_config.tensor_parallel_size
self.fd_config.model_config.tensor_parallel_rank = self.parallel_config.tensor_parallel_rank self.fd_config.model_config.tensor_parallel_rank = self.parallel_config.tensor_parallel_rank
self.fd_config.model_config.moe_group="dummy"
self.fd_config.parallel_config.column_cut = False
vision_config = self.fd_config.model_config.vision_config vision_config = self.fd_config.model_config.vision_config
vision_config.attn_sep = False vision_config.dtype = self.fd_config.model_config.dtype
vision_config.dtype = "bfloat16"
vision_config.tensor_parallel_degree = self.parallel_config.tensor_parallel_size vision_config.tensor_parallel_degree = self.parallel_config.tensor_parallel_size
vision_config.tensor_parallel_rank = self.parallel_config.tensor_parallel_rank vision_config.tensor_parallel_rank = self.parallel_config.tensor_parallel_rank
self.fd_config.model_config.pixel_hidden_size = vision_config.hidden_size
self.fd_config.model_config.im_patch_id = tokenizer.get_vocab()[ self.fd_config.model_config.im_patch_id = tokenizer.get_vocab()[
"<|IMAGE_PLACEHOLDER|>" "<|IMAGE_PLACEHOLDER|>"
] ]
self.fd_config.model_config.think_end_id = tokenizer.get_vocab()["</think>"] self.fd_config.model_config.think_end_id = tokenizer.get_vocab()["</think>"]
self.fd_config.model_config.max_text_id = self.fd_config.model_config.im_patch_id self.fd_config.model_config.sequence_parallel = self.parallel_config.sequence_parallel
self.fd_config.model_config.sequence_parallel = False
self.model_config = self.fd_config.model_config self.model_config = self.fd_config.model_config
self._init_image_preprocess() self._init_image_preprocess()
@@ -1558,9 +1553,9 @@ class GPUModelRunner(ModelRunnerBase):
rope_emb = get_rope_3d( rope_emb = get_rope_3d(
position_ids=position_ids_3d_real, position_ids=position_ids_3d_real,
rotary_dim=self.model_config.head_dim, rotary_dim=self.model_config.head_dim,
paritial_rotary_factor=1.0, partial_rotary_factor=1.0,
base=self.model_config.rope_theta, base=self.model_config.rope_theta,
max_position=self.parallel_config.max_model_len, max_position=self.parallel_config.max_model_len,
freq_allocation=self.model_config.freq_allocation, freq_allocation=getattr(self.model_config, "freq_allocation", 20),
) )
return rope_emb return rope_emb

3
fastdeploy/worker/worker_process.py
View File

@@ -567,6 +567,9 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
decoding_config = DecodingConfig(vars(args)) decoding_config = DecodingConfig(vars(args))
speculative_config = SpeculativeConfig(vars(args)) speculative_config = SpeculativeConfig(vars(args))
parallel_config = ParallelConfig(vars(args)) parallel_config = ParallelConfig(vars(args))
parallel_config.tensor_parallel_rank = local_rank
parallel_config.tensor_parallel_size = ranks
parallel_config.expert_parallel_rank = int(local_rank / ranks)
load_config = LoadConfig(vars(args)) load_config = LoadConfig(vars(args))
graph_opt_config = GraphOptimizationConfig( graph_opt_config = GraphOptimizationConfig(