funnygeeker

[funnygeeker @funnygeeker2]

@MimirLink 但是好像算上显卡折损，固态寿命消耗和电费，赚的不多的样子。

[funnygeeker @funnygeeker2]

@xueyu1125 性价比不高，思考又臭又长，速度还慢...知道能力不差但是工作中没法用(

[funnygeeker @funnygeeker2]

@BookForever1788 @xBeforeLab @wangchangfu88 对呀，就连蒲公英（植物）都被跟着限制搜不到了。就很离谱。

[funnygeeker @funnygeeker2]

@Mk_Flow 可以分享一下你的提示词增强方案吗？比如用的什么模型和什么Skill？我现在全卡在这上面。

[funnygeeker @funnygeeker2]

@Mk_Flow @hank_aibtc 等等，你还没说你的显卡代数。是50系的显卡吗？比如5060ti，因为不同代的显卡有一些支持的硬件加速不一样。

[funnygeeker @funnygeeker2]

@xBeforeLab @wangchangfu88 之前就给蒲公英路由器给封了。搞得我根本卖不了。（明明这玩意产品已经明确标明了出国就不能用了）

[funnygeeker @funnygeeker2]

@chocho111228 @Correctside_0 @cc_1022 君の中国語は本当に上手だ。

(doge)

[funnygeeker @funnygeeker2]

@__soragoto__ 不知道，反正做了UA，TTL，NTP三条防护，也算是带着几件智能家居度过了大学的三年(doge

[funnygeeker @funnygeeker2]

@1bitlabs @EmperoAI @_ryu15_ In fact, strictly speaking, the best name should be: Qwen3.6-35B-A3B-Qwen3.8-Distil

[funnygeeker @funnygeeker2]

@EmperoAI I tested it, and at present it deserves further testing.

Pay special attention, please do not turn on the `preserve_thinking` function(default value is false).
When this function is turned on, the quality of the model will be seriously reduced.

[funnygeeker @funnygeeker2]

@Summerkanna 如果你觉得回答是正常的，那你需要在问题后面加个感叹号。(๑•̀ㅂ•́)و✧

[funnygeeker @funnygeeker2]

@LingJiaXin233 B声称自己在打棒球，但身上没有雨水。下雨天根本不会去室外打棒球。因此选 B。

[funnygeeker @funnygeeker2]

@LingJiaXin233 哦，他喵的，那是绳子(・ε・｀)

[funnygeeker @funnygeeker2]

@jessyshen 我只看到了掺水的中转站(doge)

[funnygeeker @funnygeeker2]

@croudgoat 这个已经被明确不会发布了（）

[funnygeeker @funnygeeker2]

@PaulGugAI There is a model that I haven't tested yet: nex-agi/Nex-N2.5-mini

It looks worth testing, but I don't know what the final effect is.

[funnygeeker @funnygeeker2]

@naybdy76552 @PaulGugAI But I think its use as an agent is not stable. Only suitable for very short tasks.

[funnygeeker @funnygeeker2]

@sojufx @NaraharikripaM If this model is used in the work related to the browser_use(MCP). There is basically no improvement, or there is a little retrogression. But, it may perform better for the related tasks of the terminal.

[funnygeeker @funnygeeker2]

@sojufx @NaraharikripaM bro, you can try Grm3.2-Sky. I think he works well on my Hermes Agent. Although the upper limit is not as good as Ornith1.5, it has the stability of Qwen3.6.

Sometimes I feel that I am using Qwen3.7 35B-A3B. (Because there is still some gap with Qwen3.8)

[funnygeeker @funnygeeker2]

@johnny_everson @bnjmn_marie You can limit the number of thinking token to 8192. This can cut off the thinking cycle to some extent.
8192 can meet most job requirements. It is enough to simply execute 2560/4096.（But I measured 35B-A3B）