---
base_model: Qwen/Qwen3.5-9B
library_name: peft
pipeline_tag: text-generation
tags:
- verilog
- rtl
- qwen3.5
- lora
- peft
- code-generation
- repair
---

# Verilog Qwen3.5 9B v34 Manual Structured Repair LoRA

PEFT LoRA adapter for `Qwen/Qwen3.5-9B`, initialized from v33 and further trained on hand-written/manual repairs for v33 max4096 failures.

## Important data note

This adapter was trained with manual repair targets only for the targeted v33 failure repairs. No converted VerilogEval `RefModule` reference-solution code was used as repair target code.

## Intended format

```text
Thinking:
[BEGIN]
module TopModule(...);
  ...
endmodule
[DONE]
```

Use code inside `[BEGIN]` / `[DONE]` as final artifact.

## Evaluation in this project

Local VerilogEval v2 spec-to-RTL direct single-adapter eval, no retry, no selector, no compiler feedback before final answer:

```text
v34 manual max2048: compile 103/156, pass 77/156 = 49.4%
```

A max4096 eval is being run after publication request to test whether max token budget closes the gap with v33 max4096.

Interpretation: v34 manual is experimental. v33 remains the best confirmed clean single-adapter checkpoint until the max4096 v34 result is known.

## Loading

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = "Qwen/Qwen3.5-9B"
adapter = "Pablo-Flores-Mollinedo/verilog-qwen3.5-9b-v34-manual-structured-repair-lora"

tok = AutoTokenizer.from_pretrained(adapter, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base, trust_remote_code=True, device_map="auto")
model = PeftModel.from_pretrained(model, adapter)
```

## Notes

- Adapter only; requires the base model license/weights.
- Generated RTL should be compiled and simulated before use.
- Benchmark scores are finite-testbench simulation results, not formal proof.