Model Card: Xing4.0-29B-A4B-Rewrite
Model Overview
Xing4.0-29B-A4B-Rewrite is a supervised fine-tuning (SFT) on top of XingChen-AGI/Xing4.0-29B-A4B (MoE, 29B total / 4B activated). It is designed for rewriting conversational and disorganized speech transcription text into clearer and more coherent written text.
Model Description
| Field | Value |
|---|---|
| Model name | Xing4.0-29B-A4B-Rewrite |
| Base model | XingChen-AGI/Xing4.0-29B-A4B |
| Training method | Supervised Fine-Tuning (SFT) |
| Architecture | Decoder-only LLM (MoE) |
| Languages | Chinese |
| SFT context length | 2048 |
Intended Use
Typical processing includes:
- reducing unnecessary repetitions and redundant expressions;
- removing meaningless fillers and fragmented expressions when appropriate;
- repairing incomplete or grammatically broken sentences;
- reorganizing disordered sentence structures;
- converting highly colloquial expressions into more natural written language;
- avoiding unnecessary summarization or information compression.
Limitations
The model may have the following limitations:
- Ambiguous or severely fragmented transcription text may be rewritten incorrectly.
- Errors already present in the transcription may affect the rewritten result.
- Long or highly disorganized inputs may lead to local information reordering or omission.
- Performance may vary across domains, speaking styles, and transcription quality.