Add Qwen3 LLM

[sarapapi]

Description

• Model Name: Qwen3
• Model Type: LLM
• Checkpoint: https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507
• Paper/Technical Report: https://arxiv.org/abs/2505.09388
• License: Apache 2.0
• Motivation: Qwen3-Omni was added, and it's interesting to see the comparison with a cascaded version with the same LLM (Qwen3)

Outputs

List of benchmarks for which outputs are provided:

• FLEURS
• CoVoST2
• Europarl-ST
• WMT
• WinoST (Canary Patricia partial en-es, en-fr, en-it, en-pt; Whisper Sara running, Seamless Sara running)
• CommonAccent
• ManDi
• CS-Dialogue
• CS-FLEURS
• LibriStutter
• NoisyFLEURS (ambient)
• NoisyFLEURS (babble) (Canary Patricia done; OWSM Sara done; Whisper Sara done; Seamless Sara on going)
• EmotionTalk
• mExpresso (Canary Patricia partial en-fr, en-it, en-nl, en-pt, en-zh)
• ACL 60/60 (long)
• ACL 60/60 (short)
• MCIF (long)
• MCIF (short)