Qwen2可以转onnx,转bmodel的时候出现以下问题

tzhang2014 commented 2 months ago

sdk是v1.6.113-g7dc59c81-20240105, 转bmodel出现model_deploy.py: error: unrecognized arguments: --addr_mode io_alone

去掉--addr_mode 转bmodel报这个错误，我用的转模型命令是这个：./compile.sh --mode int8 --name qwen2-7b --seq_length 8192

Floating point exception (core dumped) Traceback (most recent call last): File "/workspace/tpu-mlir_v1.6.113-g7dc59c81-20240105/python/tools/model_deploy.py", line 337, in tool.build_model() File "/workspace/tpu-mlir_v1.6.113-g7dc59c81-20240105/python/tools/model_deploy.py", line 232, in build_model mlir_to_model(self.tpu_mlir, self.model, self.final_mlir, self.dynamic, File "/workspace/tpu-mlir_v1.6.113-g7dc59c81-20240105/python/utils/mlir_shell.py", line 169, in mlir_to_model _os_system(cmd) File "/workspace/tpu-mlir_v1.6.113-g7dc59c81-20240105/python/utils/mlir_shell.py", line 50, in _os_system raise RuntimeError("[!Error]: {}".format(cmd_str)) RuntimeError: [!Error]: tpuc-opt block_cache_9_bm1684x_w8bf16_final.mlir --codegen="model_file=block_cache_9.bmodel embed_debug_info=false model_version=latest" -o /dev/nu

chuxiaoyi2023 commented 2 months ago

版本太老噜，用这个https://github.com/sophgo/tpu-mlir d0cbae7

不过这个要自己编译source envsetup.sh && ./build.sh DEBUG

tzhang2014 commented 2 months ago

@chuxiaoyi2023

可以转了哈，感谢

sophgo / LLM-TPU

Qwen2可以转onnx,转bmodel的时候出现以下问题 #30