KejueAI/echo-transcriber-v0.1-step105000-ct2-float16

Echo Transcriber v0.1 Step 105000 CT2 Float16

CTranslate2/Faster-Whisper conversion of KejueAI/echo-transcriber-v0.1.

• Source adapter repo: KejueAI/echo-transcriber-v0.1
• Source revision: 020d790fee2c26a35336b6a3f1a67b36766837e0
• Source subfolder: last-checkpoint
• Trainer global step: 105000
• Base model: openai/whisper-large-v3-turbo
• Conversion date UTC: 2026-05-29T15:31:36.881297+00:00
• CTranslate2 quantization: float16

The LoRA adapter was merged into the base Whisper model with PEFT, then converted with ct2-transformers-converter for use with Faster-Whisper.

Inference Server

Set the transcriber model to:

MODEL_NAME=KejueAI/echo-transcriber-v0.1-step105000-ct2-float16
SERVED_MODEL_NAME=KejueAI/echo-transcriber-v0.1-step105000-ct2-float16
BACKEND=faster_whisper
COMPUTE_TYPE=float16

Keep STRICT_MODEL_NAME=false if clients may still send openai/whisper-large-v3-turbo in the request body.

Conversion Metadata

See echo_conversion_metadata.json in this repo.