Add support SA8797(also used by SA8397)#16695
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16695
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 21 PendingAs of commit bba3ce5 with merge base 4c8132c ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
Hello @ares89 do you intend to merge to |
I want to merge to |
a9ee422 to
da23496
Compare
|
@kirklandsign |
|
cc: @cccclai @shewu-quic @chunit-quic @haowhsu-quic @winskuo-quic since this is a QNN backend related update! |
cccclai
left a comment
There was a problem hiding this comment.
Thank you! Looks good to me, will wait to see if qcom folks have more feedback
|
Hmm somehow the CI is not triggered. Can I get some help to land this PR? @SS-JIA @Gasoonjia @JacobSzwejbka |
This PR needs a
|
Tried update branch and seems working |
|
@ares89 the lint runner is failing, can you fix the error? |
fixed ^_^ |
Summary
Fixes #16535
feat(backends/qualcomm): add support for SA8797/SA8397 chipset and V81 architecture
Test plan
I have tested the changes by compiling and running a model on the SA8397 with QNN SDK v2.42.0.251225
python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -m SA8797 --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model qwen3-0_6b --compile_only --prompt "what is 1+1" --checkpoint /home/xxx/.cache/meta_checkpoints/Qwen_Qwen3-0.6B.pth --artifact ./llama_qnn/qwen3/8797_04python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -m SA8797 -s localhost:5555 --decoder_model qwen3-0_6b --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "what is 1+1" --pre_gen_pte D:\\tmp\\llama_qnn_qwen3_06b\\8797_04cc @cccclai @winskuo-quic @shewu-quic @haowhsu-quic @DannyYuyang-quic @cbilgin