Is there any way to fix the quality degradation issue? I tried both float16 and float32, but the quality didn't improve much — though float32 did seem slightly better.
ct2-transformers-converter --model openai/whisper-large-v3 --output_dir ct2_models/whisper-large-v3-ct2_v1 --copy_files tokenizer.json preprocessor_config.json --quantization float16
ct2-transformers-converter --model openai/whisper-large-v3 --output_dir ct2_models/whisper-large-v3-ct2_v1 --copy_files tokenizer.json preprocessor_config.json --quantization float32
Is there any way to fix the quality degradation issue? I tried both float16 and float32, but the quality didn't improve much — though float32 did seem slightly better.