mlcommons / mlcommons/mlperf_client
[Bug] Phi3.5 Mini fails to initialize IHV with --temp-dir option (NVIDIA RTX 5060 Laptop & AMD Radeon 880M)
@swasson488 is already working on this.
Since Oct 9, 2025.
- Dominant language
- C++
- Stars
- 91
- Forks
- 12
- Avg merge
- 2d 19h
- Merged PRs (30d)
- 1
Description
When running Phi3.5 Mini on an NVIDIA RTX 5060 Laptop GPU (and also on AMD Radeon 880M) following Running MLPerf Client on an Offline Machine, the benchmark fails to start if the --temp-dir . option is specified. Without --temp-dir ., the benchmark runs successfully.
I'm using .\mlperf-windows.exe -c .\phi3.5\NVIDIA_ORTGenAI-DML_GPU.json --temp-dir . and I got the following:
Execution Summary:
Model: Phi3.5 phi-3.5-mini_instruct-dml (#1):
:
OrtGenAI {"device_id":0,"device_name":"NVIDIA GeForce RTX 5060 Laptop GPU","device_type":"GPU","device_vendor":"NVIDIA"}:
Error: Failed to initialize IHV,
Invalid device id: 0
Invalid device id: 0
Check Logs/error.log and Logs/phi3_5_executor.log for more details.
However, when using .\mlperf-windows.exe -c .\phi3.5\NVIDIA_ORTGenAI-DML_GPU.json (without --temp-dir .), the benchmark can run successfully.
Expected: it should work with or without --temp-dir..
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.