tCosta retweetledi
tCosta
34.3K posts
@wesleysimplic @kazzkiq Perdão pelo julgamento adiantado. Tinha entendido que era um modelo pós treinado. Aparentemente é um cli que normaliza a saída de prompt, gostei da inciativa amigo! Consigo ver utilidade no Nemotron nano que é um cocô mas possui 1M de contexto. Normalizando a saída pode ser bom!!
Português
@wesleysimplic @kazzkiq Comparativos com o qwen2.5? Ouso dizer benchmaxxing ou overfitting. Com todo respeito amigo...
Português
@kazzkiq Homi, testa ai, faz o clone, tive uma pifania ontem, veio do ceú, olha os novos testes
Português
Deepseek-v4-Flash beats Sonnet, and Opus-4.5 (no thinking) and basically matches GPT-5.2 medium
Tomorrow I will have a compression of Flash that'll make it fit well on 1x Spark at hopefully better quality than alternatives
Join tomorrow:
luma.com/reap
English
@LottoLabs Its an excellent worker agent, not so much smart to do things from scratch. My go-to is make deepseek create scripts, skills, and just make 9b follow it. Made 10usd of deepseek last a whole month this way.
English
Pfv xi ji ping nunca te pedi nada.
Red Operative Leninist@RedPill_Marxism
PC builders when they see that China is going to flood the market with cheap, powerful GPUs
Português
tCosta retweetledi
PC builders when they see that China is going to flood the market with cheap, powerful GPUs
Jackson Hinkle 🇺🇸@jacksonhinkle
🇨🇳 China Unveils Gaming GPU To Challenge NVIDIA.
English
Confirmed qwen 27b MTP works great in LMstudio
LM Studio@lmstudio
MTP is available in LM Studio 0.4.14. Sound on.
English
@codecovenant @lmstudiodevs My point being. I am using lmstudio for over a year now. The first time i complained of it over llama.cpp they released it lol. Had good result with qwen3.6 27b, on par with llamacpp. But 35b, not so much.
English
@uglesado @lmstudiodevs There have been a few of these pivots lately like dflash. Gotta test what even works best.
Lmstudio is following llama in boosting mtp.
I was waiting to see what they choose because it let's me avoid doing the homework. Easy to use their expertise .
English
LM Studio with Multi Token Prediction (MTP) is now in beta.
1. Update to 0.4.14+3 in-app
2. Make sure your llama.cpp engine is 2.15.0
3. Turn on MTP when loading a model
Use a model that supports it, like Qwen3.6-35B-A3B-MTP-GGUF or Qwen3.6-27B-MTP-GGUF
English
@brammy999 @siegnant @teortaxesTex Basically the same, the adaptations were matured long ago. When earlier adaptations in earlier 2000s, they lasted tiny bit less than mono fuel. Thats not an issue anymore.
English
wtf is Brazil doing? All their cars work on ethanol?!
John Bistline@JEBistline
The global EV transition in one chart: Leader: Norway Almost there: Netherlands, China Accelerating: U.K., France, Germany Starting: Thailand, South Korea Barely begun (yet still selling millions): U.S., India, Indonesia Doing its own thing: Brazil (ethanol)
English
@sumoto_iitoko Muito comum brasileiros torcerem para seleções estrangeiras, exceto argentina, portugal e alemanha claro...
Português
@NousResearch @Teknium Embedded calendar skill? So the agent remind the user with to-dos, reminders and personalised notes?
English
English
English
English
@uglesado @zolotukhin_ai @AiXsatoshi would recommend you to checkout something i opensourced recently
it can setup autoresearch loops for you inside codex itself
github.com/evo-hq/evo
English
M5 Max Macbook pro来たので
とりあえずcodex,claude,Lmstdioインストールして
LLMダウンロードしてる
どの量子化にするか迷いますね
日本語