model card

README.md

@@ -0,0 +1,28 @@
+---
+base_model: Qwen/Qwen3.8-27B
+library_name: peft
+license: apache-2.0
+tags: [verilog, rtl, hardware-design, lora, verilog-eval]
+---
+
+
+# RecurRTL-MATE — Qwen3.8-27B role adapters
+
+Four LoRA adapters for **Qwen/Qwen3.8-27B**, used by the RecurRTL-MATE recursive
+RTL-generation pipeline. Load one with `subfolder=`:
+
+| subfolder | role |
+|---|---|
+| `repair` | proposes a corrected module from a failing design plus its verification feedback |
+| `localizer` | identifies which part of a failing design is responsible |
+| `verifier` | judges whether a candidate design satisfies the specification |
+| `planner` | produces an implementation plan before RTL is written |
+
+```python
+model = PeftModel.from_pretrained(base, "musatur/recurrtl-mate-qwen38-adapters", subfolder="repair")
+```
+
+Base model **Qwen/Qwen3.8-27B** scores **77.56% pass@1**
+(121/156) on VerilogEval v2 before any adapter is applied.
+
+Demo: https://huggingface.co/spaces/musatur/recurrtl-mate-demo