-
Notifications
You must be signed in to change notification settings - Fork 218
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.
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
relax check on bf16 #3460
relax check on bf16 #3460
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/3460
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 245ffd8 with merge base 1fd80fe (): This comment was automatically generated by Dr. CI and updates every 15 minutes. |
@lucylq has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. @lucylq could you create an issue for bf16 kernel support? Thanks!
Summary: Now require kernel support for bf16. Export: python3 torchchat.py export llama3 --dtype [fp16 | bf16] --output-pte-path llama3.pte ``` (.venv) (base) [lfq@devvm20128.prn0 /data/users/lfq/torchchat (lfq.export-bf16)]$ python3 torchchat.py generate llama3 --device cpu --pte-path llama3.pte --prompt "Hello my name is" Warning: checkpoint path ignored because an exported DSO or PTE path specified Warning: checkpoint path ignored because an exported DSO or PTE path specified Using device=cpu Intel Core Processor (Broadwell) Loading model... Time to load model: 0.11 seconds I 00:00:00.000905 executorch:program.cpp:129] InternalConsistency verification requested but not available E 00:00:51.744419 executorch:method.cpp:936] Overriding output data pointer allocated by memory plan is not allowed. I 00:00:51.744460 executorch:pybindings.cpp:196] Cannot set_output_data_ptr(): this likely means the outputs were MemoryPlanned inspect the error code to know for sure, but likely this is not an issue. 0x2 F 00:00:51.747880 executorch:op_index.cpp:87] In function operator()(), assert failed (false): Unhandled dtype BFloat16 for index.Tensor_out Aborted (core dumped) ``` fp16 (mostly) runs well! ``` (.venv) (base) [lfq@devvm20128.prn0 /data/users/lfq/torchchat (lfq.export-bf16)]$ python3 torchchat.py generate llama3 --device cpu --pte-path llama3_f16.pte --prompt "Hello my name is" 2>/dev/null Warning: checkpoint path ignored because an exported DSO or PTE path specified Warning: checkpoint path ignored because an exported DSO or PTE path specified Using device=cpu Intel Core Processor (Broadwell) Loading model... Time to load model: 0.10 seconds Hello my name is Elsie, I'm a 28 year old woman who loves to be outdoors and try new things. I'm a bit of a thrill-seeker and love to push myself out of my comfort zone. I'm also a big fan of nature and love spending time in the wilderness. I'm a bit of a foodSerializedName and love trying new recipes and experimenting with different flavors and ingredients. I'm also a big fan of coffee and love a good cup of joe in the morning. I'm a bit of a homebody and love spending time at home with my family and friends. I'm also a big fan of movies and TV shows and love getting lost in a good story. Injector I'm a bit of a hopeless romantic and love the idea of finding true love and building a life with someone. I'm looking for someone who is kind, honest, and genuine, and who also loves the outdoors and trying new things. If you're a fellow thrill-seeker who loves nature and<|begin_of_text|>Hello my name is Elsie, I'm a 28 year old woman who loves to be outdoors and try new things. I'm a bit of a thrill-seeker and love to push myself out of my comfort zone. I'm also a big fan of nature and love spending time in the wilderness. I'm a bit of a foodSerializedName and love trying new recipes and experimenting with different flavors and ingredients. I'm also a big fan of coffee and love a good cup of joe in the morning. I'm a bit of a homebody and love spending time at home with my family and friends. I'm also a big fan of movies and TV shows and love getting lost in a good story. Injector I'm a bit of a hopeless romantic and love the idea of finding true love and building a life with someone. I'm looking for someone who is kind, honest, and genuine, and who also loves the outdoors and trying new things. If you're a fellow thrill-seeker who loves nature and<|eot_id|> Max Sequence Length Reached. Ending Conversation. ========== ``` duplication seems to be resolved in pytorch/torchchat#620 Reviewed By: iseeyuan Differential Revision: D56849464 Pulled By: lucylq
3969c01
to
245ffd8
Compare
This pull request was exported from Phabricator. Differential Revision: D56849464 |
Now require kernel support for bf16.
Export:
python3 torchchat.py export llama3 --dtype [fp16 | bf16] --output-pte-path llama3.pte
fp16 runs well!