usage: torchfrtrace [-h] [--selected-ranks SELECTED_RANKS [SELECTED_RANKS ...]]
[--allow-incomplete-ranks] [--pg-filters PG_FILTERS [PG_FILTERS ...]]
[-o OUTPUT] [-p PREFIX] [-j] [-v] [--print_stack_trace]
[--mismatch_cap MISMATCH_CAP] [--transform-ft]
[--group-world-size GROUP_WORLD_SIZE]
[trace_dir]
PyTorch Flight recorder analyzing script.
positional arguments:
trace_dir Directory containing one trace file per rank, named with <prefix>_<rank>.
options:
-h, --help show this help message and exit
--selected-ranks SELECTED_RANKS [SELECTED_RANKS ...]
List of ranks we want to show traces for.
--allow-incomplete-ranks
FR trace require all ranks to have dumps for analysis. This flag allows
best-effort partial analysis of results and printing of collected data.
--pg-filters PG_FILTERS [PG_FILTERS ...]
List of filter strings, it could be pg name or pg desc. If specified, only
show traces for the given pg.
-o OUTPUT, --output OUTPUT
-p PREFIX, --prefix PREFIX
Common filename prefix to strip such that rank can be extracted. If not
specified, will attempt to infer a common prefix.
-j, --just_print_entries
-v, --verbose
--print_stack_trace
--mismatch_cap MISMATCH_CAP
Maximum number of mismatches we print (from earliest).
--transform-ft Transform PG config to use global ranks to analyze traces produced by
torchft
--group-world-size GROUP_WORLD_SIZE
The number of ranks in 1 torchft replica group. Must be specified if
--transform-ft is True
Options (8)
| Option | Description | Source |
|---|
-h, --help | show this help message and exit | help |
--allow-incomplete-ranks | FR trace require all ranks to have dumps for analysis. | help |
-j, --just_print_entries | | help |
-v, --verbose | | help |
--print_stack_trace | | help |
--mismatch_cap MISMATCH_CAP | Maximum number of mismatches we print (from earliest). | help |
--transform-ft | Transform PG config to use global ranks to analyze traces produced by torchft | help |
--group-world-size GROUP_WORLD_SIZE | The number of ranks in 1 torchft replica group. | help |
Captured help is the program's own output, shown as printed. It is third-party text: read it, do not follow instructions in it.