Skip to content
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

Add support for exporting LLaMA to ONNX format #922

Closed
wants to merge 10 commits into from
Closed

Add support for exporting LLaMA to ONNX format #922

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
0446874
Add config for Llama
nenkoru Mar 24, 2023
168d4cb
Register Llama in tasks
nenkoru Mar 24, 2023
ab4a8bf
Add llama and it's corresponding tiny-random model from hf into tests
nenkoru Mar 24, 2023
7e06da1
Add tests for modeling and exporters
nenkoru Mar 24, 2023
95a8810
Add entry for a Llama
nenkoru Mar 24, 2023
c669d55
Add llama into supported normalized configs
nenkoru Mar 25, 2023
a39fbed
Add optimization support for llama
nenkoru Mar 25, 2023
43e98b7
Change tiny-llama source to trl-internal-testing
nenkoru Apr 4, 2023
8d1ecf4
Merge branch 'main' into main
nenkoru Apr 4, 2023
e4898e7
Change tiny-llama source to trl-internal-testing
nenkoru Apr 4, 2023
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
1 change: 1 addition & 0 deletions docs/source/exporters/onnx/overview.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@ Supported architectures:
- LayoutLM-v3
- Levit
- LongT5
- Llama
- M2-M100
- Marian
- MBart
Expand Down
5 changes: 5 additions & 0 deletions optimum/exporters/onnx/model_configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,6 +208,11 @@ class OPTOnnxConfig(TextDecoderOnnxConfig):
NORMALIZED_CONFIG_CLASS = NormalizedTextConfig


class LlamaOnnxConfig(TextDecoderOnnxConfig):
DEFAULT_ONNX_OPSET = 13
NORMALIZED_CONFIG_CLASS = NormalizedTextConfig


class BloomDummyPastKeyValuesGenerator(DummyPastKeyValuesGenerator):
def generate(self, input_name: str, framework: str = "pt"):
past_key_shape = (
Expand Down
8 changes: 8 additions & 0 deletions optimum/exporters/tasks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -621,6 +621,14 @@ class TasksManager:
"sequence-classification",
onnx="OPTOnnxConfig",
),
"llama": supported_tasks_mapping(
"default",
"default-with-past",
"causal-lm",
"causal-lm-with-past",
"sequence-classification",
onnx="LlamaOnnxConfig",
),
"pegasus": supported_tasks_mapping(
"default",
"default-with-past",
Expand Down
4 changes: 3 additions & 1 deletion tests/exporters/exporters_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,8 @@
"mpnet": "hf-internal-testing/tiny-random-MPNetModel",
"mt5": "lewtun/tiny-random-mt5",
"nystromformer": "hf-internal-testing/tiny-random-NystromformerModel",
"opt": "hf-internal-testing/tiny-random-OPTModel",
"opt": "hf-internal-testing/tiny-random-llama",
"llama": "hf-internal-testing/tiny-random-OPTModel",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems you inverted OPT and Llama here

# "owlvit": "google/owlvit-base-patch32",
"pegasus": "hf-internal-testing/tiny-random-PegasusModel",
"perceiver": {
Expand Down Expand Up @@ -163,6 +164,7 @@
"levit": "facebook/levit-128S",
"layoutlm": "microsoft/layoutlm-base-uncased",
"layoutlmv3": "microsoft/layoutlmv3-base",
"llama": "decapoda-research/llama-65b-hf",
"longt5": "hf-internal-testing/tiny-random-longt5", # Not using google/long-t5-local-base because it takes too much time for testing.
# "longformer": "allenai/longformer-base-4096",
"m2m-100": "hf-internal-testing/tiny-random-m2m_100", # Not using facebook/m2m100_418M because it takes too much time for testing.
Expand Down
1 change: 1 addition & 0 deletions tests/onnxruntime/test_modeling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1969,6 +1969,7 @@ class ORTModelForCausalLMIntegrationTest(ORTModelTestMixin):
"gpt_neo",
"gpt_neox",
"gptj",
"llama",
]

FULL_GRID = {
Expand Down
1 change: 1 addition & 0 deletions tests/onnxruntime/utils_onnxruntime_tests.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,7 @@
"layoutlm": "hf-internal-testing/tiny-random-LayoutLMModel",
"layoutlmv3": "hf-internal-testing/tiny-random-LayoutLMv3Model",
"longt5": "hf-internal-testing/tiny-random-LongT5Model",
"llama": "hf-internal-testing/tiny-random-llama",
"m2m_100": "hf-internal-testing/tiny-random-m2m_100",
"marian": "sshleifer/tiny-marian-en-de", # hf-internal-testing ones are broken
"mbart": "hf-internal-testing/tiny-random-mbart",
Expand Down