Qwen3-0.6B-ExecuTorch-XNNPACK

Qwen3-0.6B (ExecuTorch, XNNPACK, fp32)

This folder contains an ExecuTorch .pte export of Qwen/Qwen3-0.6B for CPU inference via the XNNPACK backend.

Contents

• model.pte: ExecuTorch program
• tokenizer.json, vocab.json, merges.txt: tokenizer artifacts
• config.json, generation_config.json, tokenizer_config.json: metadata files from the upstream Hugging Face repo

Export details

• Task: text-generation
• Recipe: xnnpack
• Flags: --use_custom_sdpa --use_custom_kv_cache
• Weight dtype: float32 (no quantization)

Tooling versions used to generate these artifacts:

• ExecuTorch: executorch==1.2.0a0+d265acf (git d265acfb63)
• Optimum ExecuTorch: optimum-executorch==0.2.0.dev0 (git 4c62ed77)

Command used:

optimum-cli export executorch \
  --model "Qwen/Qwen3-0.6B" \
  --task "text-generation" \
  --recipe "xnnpack" \
  --use_custom_sdpa \
  --use_custom_kv_cache \
  --output_dir "<output_dir>"

Run with the ExecuTorch Llama runner

Build the runner from the ExecuTorch repo:

make llama-cpu

Run:

cmake-out/examples/models/llama/llama_main \
  --model_path "model.pte" \
  --tokenizer_path "tokenizer.json" \
  --prompt "Simply put, the theory of relativity states that" \
  --max_new_tokens 48 \
  --temperature 0

Note: you may see an RE2 regex compilation error and a PCRE2 fallback message from the tokenizer; this is expected on some builds and does not affect correctness.