Charlie Doern 6494658a10 feat: add finetune_multi_device recipe with fsdp support ... the HF SFTTrainer supports distributed training using FSDP. Add a new recipe, `finetune_multi_device` which supports multi-GPU (cuda) training using FSDP and optionally LoRA. transformers hides _alot_ of their usage of FSDP behind the training args: a6b51e7341/src/transformers/training_args.py (L1535) you need to pass both `fsdp` and `fsdp_config` to get it to work properly. However, it seems many of the `fsdp_config` entries are silently ignored. The key things to get this working were: full_shard offload (cpu offload) transformer_layer_cls_to_wrap (model specific wrapping) cpu_ram_efficient_loading sharding_strategy limit_all_gathers sync_module_states backward_prefetch use_orig_params these can be seen both in `fsdp=` and `fsdp_config=` int he `SFTConfig` call. I have tested this with different model architectures with and without LoRA with success. the user can now toggle `recipe` in their provider config between `single` and `multi` to access the two different recipes. for debugging purposes NCCL logging settings can now be accessed via the provider config as well Signed-off-by: Charlie Doern <cdoern@redhat.com>		2025-06-12 13:33:33 -04:00
..
inline	feat: add finetune_multi_device recipe with fsdp support	2025-06-12 13:33:33 -04:00
registry	feat: add deps dynamically based on metastore config (#2405)	2025-06-05 14:07:25 -07:00
remote	fix(weaviate): handle case where distance is 0 by setting score to infinity (#2415)	2025-06-12 11:23:59 -04:00
utils	fix: set appropriate defaults for params (#2434)	2025-06-11 17:30:34 -07:00
__init__.py	API Updates (#73)	2024-09-17 19:51:35 -07:00
datatypes.py	fix(tools): do not index tools, only index toolgroups (#2261)	2025-05-25 13:27:52 -07:00