From 9b4e9cc8a0311e5243d69b73ed073e7ea441982e Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 27 三月 2024 16:05:29 +0800
Subject: [PATCH] train update
---
examples/industrial_data_pretraining/llm_asr/conf/whisper_vicuna_linear.yaml | 16 ++++++++--------
1 files changed, 8 insertions(+), 8 deletions(-)
diff --git a/examples/industrial_data_pretraining/llm_asr/conf/whisper_vicuna_linear.yaml b/examples/industrial_data_pretraining/llm_asr/conf/whisper_vicuna_linear.yaml
index eacbd35..8a01d14 100644
--- a/examples/industrial_data_pretraining/llm_asr/conf/whisper_vicuna_linear.yaml
+++ b/examples/industrial_data_pretraining/llm_asr/conf/whisper_vicuna_linear.yaml
@@ -58,26 +58,26 @@
train_conf:
accum_grad: 1
grad_clip: 5
- max_epoch: 150
+ max_epoch: 15
keep_nbest_models: 10
log_interval: 10
optim: adamw
optim_conf:
lr: 0.0001
- weight_decay: 0.000001
-scheduler: warmuplr
+ weight_decay: 0
+
+scheduler: custom_lambdalr
scheduler_conf:
- warmup_steps: 1500
+ warmup_steps: 1000
dataset: AudioLLMVicunaDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: RankFullLocalShuffleBatchSampler
+ batch_sampler: CustomDistributedBatchSampler
batch_type: example # example or length
- batch_size: 8 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
- max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
- buffer_size: 500
+ batch_size: 4 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
+ max_token_length: 3000 # filter samples if source_token_len+target_token_len > max_token_length,
shuffle: True
num_workers: 4
# preprocessor_text: TextPreprocessRemovePunctuation
--
Gitblit v1.9.1