mirror of
https://github.com/huggingface/accelerate.git
synced 2025-10-20 18:13:46 +08:00
8b55e62b2c
* xpu gather issue fixed in 2.9 and validated config_yamls on XPU Signed-off-by: YAO Matrix <matrix.yao@intel.com> * xxx Signed-off-by: YAO Matrix <matrix.yao@intel.com> --------- Signed-off-by: YAO Matrix <matrix.yao@intel.com>
16 lines
828 B
YAML
16 lines
828 B
YAML
# Similar to FSDP, we set the distributed type as DEEPSPEED
|
|
distributed_type: DEEPSPEED
|
|
# With DeepSpeed, we utilize a deepspeed config file for the entire configuration
|
|
deepspeed_config:
|
|
# Can also be any of the config json's in accelerate/examples/deepspeed_config_templates
|
|
deepspeed_config_file: ../deepspeed_config_templates/zero_stage1_config.json
|
|
# If using ZeRO-3 and wanting to load big models in, this should be set to `true` so
|
|
# `transformers` uses the right `init` function
|
|
zero3_init_flag: false # true
|
|
|
|
# Finally we need to specify the number of accelerators to use
|
|
num_processes: 2
|
|
# Optionally we can set the mixed precision now instead of in the deepspeed config file,
|
|
# however this requires the `fp16` and `bf16` options to be set to `auto` in the deepspeed config file
|
|
# mixed_precision: "bf16"
|