DeepRec
DeepRec copied to clipboard
DeepRec is a high-performance recommendation deep learning framework based on TensorFlow. It is hosted in incubation in LF AI & Data Foundation.
export TF_USE_CUBLASLT=1 got
**System information** - DeepRec version (you are using): 1.15.5+deeprec2208 - Are you willing to contribute it (Yes/No): Yes **Describe the feature and the current behavior/state.** how deeprec analyze the sparse...
**System information** - OS Platform and Distribution (e.g., Linux Ubuntu 16.04): - DeepRec version or commit id: 1.15.5+deeprec2208 - Python version: python3.6 - Bazel version (if compiling from source): bazel...
**System information** - OS Platform and Distribution (e.g., Linux Ubuntu 20.04): Ubuntu 20.04 - DeepRec version or commit id: 23252970 - Python version: python3.6.9 - Bazel version (if compiling from...
### 训练期间ps的cpu使用率变化情况  ### 训练期间chief的cpu使用率变化情况(worker类似)  ### 训练期间每秒训练批次变化情况  tf1.15版本,使用1chief 1ps 4worker进行分布式训练,训练期间ps的cpu持续增长,chief和worker的cpu后续有降低的情况,每秒训练批次也变少了,这是因为什么原因?
Evict支持scatter_update操作吗 ev_opt = tf.EmbeddingVariableOption(init_option=init_opt, filter_option=filter_opt, evict_option=evict_opt) emb_table = tf.get_embedding_variable("ev_emb_table", embedding_dim=64, partitioner=tf.fixed_size_partitioner(num_shards=10), ev_option=ev_opt) emb_table能像普通embedding_table一样支持scatter_update操作吗
**System information** - DeepRec version (you are using): - Are you willing to contribute it (Yes/No):yes **Describe the feature and the current behavior/state.** 推荐一般需要回溯很多天的数据,进行eval/train 流式训练,可以提供多天流式训练的demo例子吗? 例如这种: https://github.com/PaddlePaddle/PaddleRec/blob/master/doc/online_trainer.md **Will this change...
你好, 调试中遇到一个问题, 有些疑问, 如果方便, 辛苦帮忙解答. 感谢 我在测试kafkadataset的时候, 会产生非常多线程. 但有利用率的占比不高. 随着partition数的增多, 线程数成比例增加. (1) 单机情况下, 200个partition, 可以产生超过15000个线程. (dataset设置的并行度是3, inter_op, intra_op都默认, 在128core, 64core, 32core机器测试结果都一样线程数一样多). (2) 分布式情况下, 2个worker, 各assign 100个partition, 单worker线程数减半(7-8k个), 总数不变 同理, 扩展到4个,...