Views
No views yet
iverilog/vvp feedback.1base: Qwen/Qwen3.5-9B
2primary adapter: Pablo-Flores-Mollinedo/verilog-qwen3.5-9b-v33-thinking-reinforced-lora
3repair adapter: Pablo-Flores-Mollinedo/verilog-qwen3.5-9b-v34-manual-structured-repair-lora
4verifier: iverilog + vvp
5max attempts: 5 model calls/problem1benchmark: VerilogEval v2 spec-to-RTL
2total: 156
3compile: 145/156 = 92.95%
4pass: 95/156 = 60.90%
5total model calls: 464
6avg calls/problem: 2.971attempt 1: 58
2attempt 2: 17
3attempt 3: 14
4attempt 4: 5
5attempt 5: 11v33 clean single adapter: pass 80/156
2v35 long-budget single adapter: pass 77/156
3v29 previous practical pipeline: pass 84/156
4v36 verified agentic pipeline: pass 95/156VerilogEval v2 verified agentic pipeline, max B=5 model calls/problem, pass 95/156.1single-adapter pass@1
2zero-shot single-model result1scripts/run_v36_agentic_verified_pipeline.py
2scripts/run_v36_agentic_verified_pipeline.sh
3results/direct_summary.json
4results/direct_summary_results.jsonl
5results/v36_pipeline_config.json1MAX_ATTEMPTS=5 \
2MAX_NEW_TOKENS=4096 \
3REPAIR_MAX_NEW_TOKENS=4096 \
4BUILD=/home/pablo/verilog-eval/build-v36-qwen35-v33-v34-agentic-verified-maxB5 \
5bash scripts/run_v36_agentic_verified_pipeline.sh