Process killed (OOM?) when compiling Qwen3-30B-A3B

[lifelongeeek]

Describe the bug
I tried to compile and test inference of Qwen3-30B-A3B-Instruct-2507(which is listed in validated model), but process killed (possibly due to OOM).

To Reproduce
Steps to reproduce the behavior:

1. Command Used to run / script used
2. Error details

(qeff_env) ai@ai-ABPI-130:~/workspace/gmkim/efficient-transformers$ python -m QEfficient.cloud.infer --model_name Qwen/Qwen3-30B-A3B-Instruct-2507  --batch_size 1 --prompt_len 32 --ctx_len 128 --mxfp6 --num_cores 24 --device_group [0] --prompt "My name is" --mos 1 --aic_enable_depth_first
/home/ai/workspace/gmkim/qeff_env/lib/python3.10/site-packages/onnxscript/converter.py:816: FutureWarning: 'onnxscript.values.Op.param_schemas' is deprecated in version 0.1 and will be removed in the future. Please use '.op_signature' instead.
  param_schemas = callee.param_schemas()
WARNING - QEfficient - mxfp6 is going to be deprecated in a future release, use -mxfp6_matmul instead.
[Warning]: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
Loading checkpoint shards: 100%|████████████████████████████████████████████████████████████████| 16/16 [00:38<00:00,  2.41s/it]
Killed

Expected behavior
Model compiled and inference proceed

Screenshots

Environment (please complete the following information):

• OS: Linux ai-ABPI-130 6.8.0-86-generic 87~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Mon Sep 29 09:48:07 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
• Environment details with packages version etc.
• Version/Branch/Commit ID [e.g. 22]: v1

Additional context
Add any other context about the problem here.

[anujgupt-github]

@lifelongeeek : Can you please share details on the system RAM? the error seems to be due to higher RAM requirement of ONNX export.
Also, do share the AI100 SW SDK version you are using

[lifelongeeek]

I have around 186GB free system RAM.

$free -h

               total        used        free      shared  buff/cache   available
Mem:           188Gi       1.1Gi       186Gi        20Mi       527Mi       185Gi
Swap:          2.0Gi       558Mi       1.5Gi

If ONNX export is the problem, is there any workaround that I can start with pre-exported ONNX or QNN model to test inference of Qwen3-MoE model?

[lifelongeeek]

AI100 SW SDK version is probably 0.0.1.dev0.

>>> import QEfficient
>>> QEfficient.__version__
'0.0.1.dev0'

[anujgupt-github]

186 GB RAM may be less for this model export. Pre-exported model's aren't available from Qualcomm. You may try adding swap memory from disk.
How are you accessing the AI100 HW ?