[BUG] File bug report[SA8775 ADP][Qwen2.5-1.5B-Instruct] QNN context binary compile fails: Conv2d op validation failure (“BQ is not supported”)

[Undefinedx-Zhang]

I’m trying to deploy Qwen2.5-1.5B-Instruct to SA8775 ADP using AI Hub Models / Workbench pipeline. I quantified the model with w4（weight Int 4 & activation fp16).The compilation step fails during conversion to QNN context binary with an HTP backend op validation error on a Conv2d node in the attention Q projection (SHA path).

detailed error：
Failure Reason

Conversion to context binary failed with exit code 14

Additional Information from the Compile Log
Error occurred due to op : _model_model_layers_0_self_attn_q_proj_sha_0_Conv;
op_type : Conv2d;
reason : BQ is not supported
Failed to validate op _model_model_layers_0_self_attn_q_proj_sha_0_Conv with error 0xc26 : QNN_OP_PACKAGE_ERROR_VALIDATION_FAILURE
For additional information please refer to the QNN HTP Backend Op Definition Supplement : https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-10/HtpOpDefSupplement.html

[mestrona-3]

Hi @Undefinedx-Zhang , apologies for the delayed reply. We're working with the QNN HTP team on resolving this error which is affecting w4fp16 models on Auto chipsets. That said, we don't currently support Qwen2.5-1.5B on SA8775 ADP. We're working on validating various LLMs reliably on Auto chipsets and will share more via Slack when we can. Thanks.