Method and device of training deep learning recommendation model
AssigneeSamsung Electronics Co., Ltd.
InventorsHaonan Feng, Kaige Ma, Yutao Li, Jun Kim, Seungwoo Lim, Yeonjae Kim
Life of this patent
Estimating the term…
Abstract
Methods, devices, and systems for training a deep learning recommendation model (DLRM), including: offloading one or more storage tables to a memory device; training a DLRM based on training data, wherein during the training, a storage table prefetched from the memory device is loaded according to a training phase; and storing feature data obtained during the training into the loaded storage table.
Claims33
A method for training a deep learning recommendation model (DLRM), the method being performed by a first processor, and comprising:
- offloading one or more storage tables to a memory device;
- training a DLRM based on training data, wherein during the training, a storage table prefetched from the memory device is loaded according to a training phase;
- and
- storing feature data obtained during the training into the loaded storage table.
A method for training a deep learning recommendation model (DLRM), the method being performed by a second processor, and comprising:
- prefetching a storage table from a memory device according to a training phase associated with training of a DLRM by a first processor, wherein the storage table is configured to store feature data obtained during the training;
- and
- based on the storage table being prefetched, transferring the prefetched storage table to the first processor.
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
(canceled)
A system for training a deep learning recommendation model (DLRM), the system comprising:
- a memory device;
- a first processor configured to:
- offload one or more storage tables to the memory device,
- train a DLRM based on training data, and
- load a storage table prefetched from the memory device according to a training phase, during the training of the DLRM;
- and
- a second processor configured to:
- prefetch the storage table from the memory device according to the training phase, and
- based on the storage table being prefetched, transfer the prefetched storage table to the first processor,
- wherein the first processor is further configured to store feature data obtained during the training into the loaded storage table.
(canceled)
(canceled)
Dependencies are inferred from the claim text.