Atom Electron 1.4-9B LoRA
Production LoRA adapter trained from Qwen/Qwen3.5-9B on the reviewed
multitask-v3 corpus. This repository contains adapter weights and processor
artifacts, not a merged full-precision base model.
Training summary
- Training examples: 710
- Validation examples: 32
- Epochs: 1.0
- Learning rate: 2e-05
- LoRA rank/alpha: 16/16
- Maximum sequence length: 32768
- Validation loss: 0.5833584070205688
- Intended repository visibility: private
Use with the exact base model named in adapter_config.json. Evaluate the
adapter before merging or replacing a production endpoint.
- Downloads last month
- 19