From 7357bb6c45eb3f0b4d2be5a4b025385ff9eec9e5 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 21 二月 2024 14:34:36 +0800
Subject: [PATCH] update train recipe
---
funasr/bin/train.py | 17 ++++++++++-------
1 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/funasr/bin/train.py b/funasr/bin/train.py
index 8ea0c0d..0661452 100644
--- a/funasr/bin/train.py
+++ b/funasr/bin/train.py
@@ -1,3 +1,6 @@
+#!/usr/bin/env python3
+# -*- encoding: utf-8 -*-
+
import os
import sys
import torch
@@ -41,14 +44,16 @@
def main(**kwargs):
print(kwargs)
+
# set random seed
- tables.print()
set_all_random_seed(kwargs.get("seed", 0))
torch.backends.cudnn.enabled = kwargs.get("cudnn_enabled", torch.backends.cudnn.enabled)
torch.backends.cudnn.benchmark = kwargs.get("cudnn_benchmark", torch.backends.cudnn.benchmark)
torch.backends.cudnn.deterministic = kwargs.get("cudnn_deterministic", True)
local_rank = int(os.environ.get('LOCAL_RANK', 0))
+ if local_rank == 0:
+ tables.print()
# Check if we are using DDP or FSDP
use_ddp = 'WORLD_SIZE' in os.environ and int(os.environ["WORLD_SIZE"]) > 1
use_fsdp = kwargs.get("use_fsdp", None)
@@ -76,9 +81,8 @@
frontend = frontend_class(**kwargs["frontend_conf"])
kwargs["frontend"] = frontend
kwargs["input_size"] = frontend.output_size()
-
- # import pdb;
- # pdb.set_trace()
+
+
# build model
model_class = tables.model_classes.get(kwargs["model"])
model = model_class(**kwargs, **kwargs["model_conf"], vocab_size=len(tokenizer.token_list))
@@ -144,9 +148,8 @@
# dataset
dataset_class = tables.dataset_classes.get(kwargs.get("dataset", "AudioDataset"))
- dataset_tr = dataset_class(kwargs.get("train_data_set_list"), frontend=frontend, tokenizer=tokenizer, **kwargs.get("dataset_conf"))
- dataset_val = dataset_class(kwargs.get("valid_data_set_list"), frontend=frontend, tokenizer=tokenizer,
- **kwargs.get("dataset_conf"))
+ dataset_tr = dataset_class(kwargs.get("train_data_set_list"), frontend=frontend, tokenizer=tokenizer, is_training=True, **kwargs.get("dataset_conf"))
+ dataset_val = dataset_class(kwargs.get("valid_data_set_list"), frontend=frontend, tokenizer=tokenizer, is_training=False, **kwargs.get("dataset_conf"))
# dataloader
batch_sampler = kwargs["dataset_conf"].get("batch_sampler", "DynamicBatchLocalShuffleSampler")
--
Gitblit v1.9.1