You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: modelzoo/features/pmem/README.md
+3-3Lines changed: 3 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -26,15 +26,15 @@ pandas 1.1.5
26
26
--num_mock_cols=30
27
27
--batch_size=12800 \
28
28
--dim_size=256 \
29
-
--ev_storage=pmem \
29
+
--ev_storage=pmem_libpmem \
30
30
31
31
## use criteo data
32
32
./launch.sh \
33
33
--use_mock_data=False \
34
34
--data_dir=${path_to_criteo} \
35
35
--batch_size=12800 \
36
36
--dim_size=256 \
37
-
--ev_storage=pmem \
37
+
--ev_storage=pmem_libpmem \
38
38
```
39
39
40
40
- Here `num_mock_cols` is the number of columns in mock training data, `mock_vocabulary_size` is
@@ -44,5 +44,5 @@ would increase the `vocabulary_size` of each column (variable) by a factor of 2.
44
44
- The log files are located at `./bench-ps.log` and `./bench-worker.log` for ps and worker process, respectively.
45
45
- The default embedding variable type is hash table-based Embedding Variable, which does not require a specification on `mock_vocabulary_size` and `vocabulary_amplify_factor`
46
46
- by specifying `--use_ev_var=False`, the benchmark would fallback to use TF's native variable for Embeddings, and users shall provide `mock_vocabulary_size` in mock data scenarios.
47
-
- '--ev_storage=pmem/dram' to select EmbeddingVariable StorageType, Default is 'dram'.
47
+
- '--ev_storage=pmem_libpmem/pmem_memkind/dram' to select EmbeddingVariable StorageType, Default is 'dram'. If set pmem_libpmem, please also set pmem path and size with '--ev_storage_path=<pmem_path (default: /mnt/pmem0/allocator/)> and --ev_storage_size_gb=<pmem_size (default: 512)>'
0 commit comments