Deploy and benchmark Qwen3-ASR models with local, vLLM, or DashScope inference.
npx skills add https://github.com/surus-lat/benchy --skill qwen3-asr-howto
Please help me install this Agent Skill. Skill: qwen3-asr-howto Source: https://github.com/surus-lat/benchy/tree/main/.agent/skills/qwen3-asr-howto Command: npx skills add https://github.com/surus-lat/benchy --skill qwen3-asr-howto
This skill resolves the complex environment and dependency conflicts associated with running Qwen3-ASR models, ensuring they can be successfully integrated into benchmarking workflows alongside other speech models.
Run the qwen3-asr-howto skill to configure your environment and execute a smoke test for the Qwen3-ASR-0.6B model.
High-intent search queries and answers about installing and using this skill.
To run Qwen3-ASR locally, you must configure an isolated Python virtual environment to manage transformers version conflicts with other speech models. This dedicated setup ensures your existing installations remain intact while enabling local inference.
Deploying Qwen3-ASR causes dependency conflicts because it requires a specific transformers library version that clashes with other tools. Setting up an isolated Python virtual environment resolves these library dependency conflicts for benchmarking workflows.
Yes, you can use vLLM backend acceleration to speed up Qwen3-ASR inference on local hardware. The skill provides guidance for vLLM acceleration alongside local Python implementation and cloud-based DashScope integration.
The best way to benchmark Qwen3-ASR multilingual speech recognition is by using an isolated environment to run inference on FLEURS datasets. This approach allows accurate performance measurement without disrupting existing speech model installations.
Yes, the Qwen3-ASR deployment supports cloud API usage through DashScope integration. This provides an alternative to local inference and vLLM acceleration for processing speech-to-text tasks across diverse hardware environments.
You can execute a smoke test for the Qwen3-ASR-0.6B model by running the skill after configuring your isolated Python environment. This validates your local inference setup and verifies model functionality.
tachyon-beepSelect attention mechanisms to reduce memory use for long sequences.
surus-latBenchmark Whisper-family ASR models on FLEURS datasets with WER and CER metrics.
cinienceValidate the request template and connection for Model Studio Qwen ASR Realtime.
surus-latBenchmark local ASR models on FLEURS datasets with WER and CER tables.
