Qwen3.8-9B (Empero)
by Empero AI · Model · Language & code models
Community distillation of Qwen3.8 2.4T A95B into a 9B model, trained full-parameter on 70,000 teacher reasoning traces
AccessOpen weightsNeeds~44 GBSize9B parametersLicenseApache-2.0
Evidence
- Aug 17Trained with cross-entropy distillation on teacher text, not logit distillation, as Qwen3.8 log probs were unavailable
- Aug 17Built a working multi-service Docker crypto tracker (API, frontend, Redis, Nginx) from one agentic prompt; Solana data stayed empty