apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

support vl ori_vacab_size (#2900)

Browse Source
This commit is contained in:
gaoziyuan
2025-07-18 16:26:14 +08:00
committed by GitHub
parent d306944f4f
commit 6efad14b95
6 changed files with 164 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/worker/worker_process.py
+3 -4
View File
@@ -25,7 +25,8 @@ import paddle.distributed.fleet as fleet
from fastdeploy.config import (DecodingConfig, DeviceConfig, FDConfig,
GraphOptimizationConfig, LoadConfig,
ModelConfig, ParallelConfig, SpeculativeConfig)
ModelConfig, ParallelConfig, SpeculativeConfig,
ErnieArchitectures)
from fastdeploy.input.ernie_tokenizer import ErnieBotTokenizer
from fastdeploy.inter_communicator import EngineWorkerQueue as TaskQueue
from fastdeploy.inter_communicator import IPCSignal
@@ -653,9 +654,7 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
quant_config_name = args.quantization
quantization_config["quantization"] = quant_config_name
# Special handling for Ernie models
is_ernie = "Ernie4_5_ForCausalLM" in model_config.architectures or \
"Ernie4_5_MoeForCausalLM" in model_config.architectures or \
"Ernie4_5_VLMoeForConditionalGeneration" in model_config.architectures
is_ernie = ErnieArchitectures.contains_ernie_arch(model_config.architectures)
if quant_config_name == "wint4" and is_ernie:
quantization_config["dense_quant_type"] = "wint8"
quantization_config["moe_quant_type"] = "wint4"