Skip to content

Disable SDPA Attention for gpt-bigcode model #78

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
libinta merged 2 commits into from
Mar 5, 2024
Merged

Disable SDPA Attention for gpt-bigcode model #78

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ef9abea
Disable SDPA Attention for gpt-bigcode model
jiminha Feb 27, 2024
9ac19b0
Update argument to take "attn_implementation"
jiminha Mar 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion examples/text-generation/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,8 @@ python ../gaudi_spawn.py --use_deepspeed --world_size 8 run_generation.py \
> --use_hpu_graphs \
> --use_kv_cache \
> --max_new_tokens 100 \
> --bf16
> --bf16 \
> --attn_implementation eager
> ```


Expand Down
7 changes: 7 additions & 0 deletions examples/text-generation/run_generation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,13 @@ def setup_parser(parser):
action="store_true",
help="Whether to enable device map auto. In case no space left on cpu, weights will be offloaded to disk.",
)
parser.add_argument(
"--attn_implementation",
type=str,
help={"Choose whether to override framework configuration to use torch scale dot product attention or not. Note this is not same as HPU FusedSDPA."},
choices= ["eager", "sdpa"],
)

args = parser.parse_args()

if args.torch_compile:
Expand Down
3 changes: 3 additions & 0 deletions examples/text-generation/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -379,6 +379,9 @@ def initialize_model(args, logger):
model_kwargs["device_map"] = "auto"
model_kwargs["offload_folder"] = "/tmp/offload_folder/"

if args.attn_implementation:
model_kwargs["attn_implementation"] = args.attn_implementation

model = (
setup_model(args, model_dtype, model_kwargs, logger)
if not use_deepspeed
Expand Down