xiaolesu/Qwen3-8B-Herald-SFT
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Feb 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

The xiaolesu/Qwen3-8B-Herald-SFT is an 8 billion parameter QLoRA fine-tune of Qwen/Qwen3-8B, developed by xiaolesu / Osmosis AI. This causal language model specializes in natural-language-to-Lean-4 autoformalization, translating English mathematical statements into formal Lean 4 theorem stubs. It is specifically designed as a supervised cold-start checkpoint for reinforcement learning, generating structured JSON output for downstream GRPO trainers.

Loading preview...