adapter-attached-instruction-tuned
Standalone BF16 model obtained by merging the final counseling SFT LoRA adapter
into Qwen/Qwen3.8-27B.
The adapter continued DAPT training with counseling SFT.
The final adapter is merged into the model weights; a separate PEFT adapter is
not required. The complete model includes the original vision and MTP weights,
tokenizer, chat template, processor configurations, and generation configuration.
- Architecture:
Qwen3_5ForConditionalGeneration
- Precision: BF16
- Weight shards: 16 safetensors files, approximately 55.6 GB
- Merge environment: Transformers 5.19.0.dev0, PEFT 0.21.2, PyTorch 2.10.0+cu128
See merge_info.json for adapter settings and merge verification details.
The Apache 2.0 license is included in LICENSE.