Isaac Fino
Isaac Fino
暗牧策略问题
在2025/02月度的狂野排位中使用这套卡组进行自动策略测试的时候发现的些许严重的策略错误，至少因此输掉了30%的对局（粗略估计）: e.g. 1. 会无视牧师卡牌：迪菲亚麻风侏儒的战吼触发条件，优先拍出暗影法术（如，亡者复生） e.g. 2. 在不该解场的时候解场——暗牧是1-3费用铺场，后续直伤的卡组，理论上不应该在任何时候用英雄技能攻击随从（当然存在例外，比如说内战的时候考虑是否需要正好用英雄技能点掉对面的宝藏经销商，抑或是不得不优先解决嘲讽随从，等）；常常会出现回合开始先用英雄技能点击对方3血怪然后留着没有随从交互。 e.g. 3 开局留牌机制问题，过分追求全胡（3费站满海盗），没有考虑保留当前能够打出去的牌的长久之计，会出现：3张牌（保留狂暴邪翼蝠）的情况，这是很败笔的。 当然很大的一部分原因是排位匹配存在刻意压胜率的问题，因此脚本使用时间长了之后明显感觉到以上的问题会增多。 希望得到repo主的回复=w=/
When there is a VLLM serving error, such as out of the length of the max length of a tiny LLM, the output is empty, then the trace backward to...
An error happens when we execute: ```python import os from agentlightning import Trainer, DevTaskLoader, LLM from calc_agent import CalcAgent def dev_task_loader() -> DevTaskLoader: return DevTaskLoader( tasks=[ { "question": "What is...
Hello! Thank you for curating this excellent list. I'd like to suggest adding **AgentFlow** to this collection. *AgentFlow* introduces a trainable multi-agent framework that coordinates four specialized modules (planner, executor,...
Hello! Thank you for curating this excellent list. I'd like to suggest adding **AgentFlow** to this collection. *AgentFlow* introduces a trainable multi-agent framework that coordinates four specialized modules (planner, executor,...
Hello! Thank you for curating this excellent list. I'd like to suggest adding **AgentFlow** to this collection. *AgentFlow* introduces a trainable multi-agent framework that coordinates four specialized modules (planner, executor,...
Hello! Thank you for curating this excellent list. I'd like to suggest adding **AgentFlow** to this collection. *AgentFlow* introduces a trainable multi-agent framework that coordinates four specialized modules (planner, executor,...