satoyutaka/Qwen3-4B-AgentBench-llm2025_advance_v5_ironguard-fp8

Qwen3-4B-AgentBench-llm2025_advance_v5_ironguard-speed-fp8

This is the Speed-Optimized version of the V5 "Iron Guard" Agent. Generated by converting the V5-BF16 weights to FP8 (Float8) to overcome high latency per iteration.

🛡️ "Iron Guard" Discipline + Speed

This version maintains extreme discipline and strict Action format enforcement while reducing model size by 50% (4GB) for faster inference.

[日本語訳] 本モデルは、SQL等の推論遅延を解消するために、V5 (Iron Guard) を FP8 量子化 した高速版です。規律を削ることなく、軽量化（4GB）によってスループットを向上させています。