#unsloth — Public Fediverse posts

🚀✨ Look, it's 2026 and apparently, #Unsloth and #Nvidia are on a mission to squeeze every last drop of speed from GPUs; as if anyone out there was asking for yet another way to melt their consumer-grade hardware. 🤯 The authors—who clearly have more names than followers—promise #efficiency gains that’ll make you wonder why you ever settled for only 75% of your LLM training speed in the first place. 🙃
https://unsloth.ai/blog/nvidia-collab #GPUs #LLMTraining #TechNews #HackerNews #ngated

🚀✨ Look, it's 2026 and apparently, #Unsloth and #Nvidia are on a mission to squeeze every last drop of speed from GPUs; as if anyone out there was asking for yet another way to melt their consumer-grade hardware. 🤯 The authors—who clearly have more names than followers—promise #efficiency gains that’ll make you wonder why you ever settled for only 75% of your LLM training speed in the first place. 🙃
https://unsloth.ai/blog/nvidia-collab #GPUs #LLMTraining #TechNews #HackerNews #ngated

🚀✨ Look, it's 2026 and apparently, #Unsloth and #Nvidia are on a mission to squeeze every last drop of speed from GPUs; as if anyone out there was asking for yet another way to melt their consumer-grade hardware. 🤯 The authors—who clearly have more names than followers—promise #efficiency gains that’ll make you wonder why you ever settled for only 75% of your LLM training speed in the first place. 🙃
https://unsloth.ai/blog/nvidia-collab #GPUs #LLMTraining #TechNews #HackerNews #ngated

🚀✨ Look, it's 2026 and apparently, #Unsloth and #Nvidia are on a mission to squeeze every last drop of speed from GPUs; as if anyone out there was asking for yet another way to melt their consumer-grade hardware. 🤯 The authors—who clearly have more names than followers—promise #efficiency gains that’ll make you wonder why you ever settled for only 75% of your LLM training speed in the first place. 🙃
https://unsloth.ai/blog/nvidia-collab #GPUs #LLMTraining #TechNews #HackerNews #ngated

🚀✨ Look, it's 2026 and apparently, #Unsloth and #Nvidia are on a mission to squeeze every last drop of speed from GPUs; as if anyone out there was asking for yet another way to melt their consumer-grade hardware. 🤯 The authors—who clearly have more names than followers—promise #efficiency gains that’ll make you wonder why you ever settled for only 75% of your LLM training speed in the first place. 🙃
https://unsloth.ai/blog/nvidia-collab #GPUs #LLMTraining #TechNews #HackerNews #ngated

How Unsloth and Nvidia made LLM training 25% faster on consumer GPUs

https://unsloth.ai/blog/nvidia-collab

#HackerNews #Unsloth #Nvidia #LLMtraining #ConsumerGPUs #AItechnology

我的顯卡是8G的a2000，記億體好似是2600，因為不支持超頻跑不到最快速度，我用的llamacpp還沒有turbo kv cache，也沒法同時啟用no-mmap和mlock，應該是mlock有問題會crush，結果能跑到23+ tokens per second，完全是可用狀態，模型是 #unsloth 的 #qwen 3.6 35b a3b udq4km

https://youtu.be/8F_5pdcD3HY?si=jGt3qqUW82uVWeFs

[Перевод] Локальный запуск GLM-5.1

Перевод подготовил автор канала Друг Опенсурса , приятного прочтения, заранее благодарю за подписку В этой статье мы подробно разберем процесс развертывания GLM-5.1 с использованием llama.cpp и форматов GGUF. Узнаем о системных требованиях, сборке и настройках, оптимизации и практическом применении.

https://habr.com/ru/articles/1022242/

#glm51 #llm #Llamacpp #Unsloth #GGUF #Локальный_запуск #tool_calling #Zai #искусственный_интеллект

Джентльменский набор LLM-инженера: гайд по экосистеме языковых моделей

Каждый, кто хоть раз вводил pip install transformers , наблюдал, как терминал начинает безостановочно выводить простыню зависимостей: pytorch , accelerate , bitsandbytes , peft и многие, многие другие. Но если PyTorch является фундаментом, настоящим Атлантом, на плечах которого держатся тензорные вычисления, то какую роль играют его помощники? В этой статье мы проведём ревизию джентльменского набора LLM инженера. Для этого мы изучим функционал, методы работы и даже заглянем в исходный код таких библиотек, как PyTorch, Transformers, Accelerate, Bitsandbytes, PEFT и Unsloth. Эти знания позволят вам видеть за списком импортов не просто названия, а четкую структуру, на которой держится ваше приложение.

https://habr.com/ru/articles/984248/

#LLMэкосистема #pytorch #accelerate #transformers #bitsandbytes #peft #unsloth #распределённое_обучение #граф_вычислений #квантование

I am testing the capabilities of some small #LLM 's on #LMStudio today. These 7 to 12 B models are much stronger than I thought. Some of them run pretty fast, but some larger models are burning my #rtx4060 #GPU. I think I will settle with #IBM #Granite 3.3 which is a 8B model but was further trained by #unsloth to 9B. Granite 3.3 came out in April this year. In the long run, I will need a 20 to 40B model. But then I most likely need an rtx 5090 machine with 64G VRAM to run them. #AI #AIs

I am testing the capabilities of some small #LLM 's on #LMStudio today. These 7 to 12 B models are much stronger than I thought. Some of them run pretty fast, but some larger models are burning my #rtx4060 #GPU. I think I will settle with #IBM #Granite 3.3 which is a 8B model but was further trained by #unsloth to 9B. Granite 3.3 came out in April this year. In the long run, I will need a 20 to 40B model. But then I most likely need an rtx 5090 machine with 64G VRAM to run them. #AI #AIs

I am testing the capabilities of some small #LLM 's on #LMStudio today. These 7 to 12 B models are much stronger than I thought. Some of them run pretty fast, but some larger models are burning my #rtx4060 #GPU. I think I will settle with #IBM #Granite 3.3 which is a 8B model but was further trained by #unsloth to 9B. Granite 3.3 came out in April this year. In the long run, I will need a 20 to 40B model. But then I most likely need an rtx 5090 machine with 64G VRAM to run them. #AI #AIs

I am testing the capabilities of some small #LLM 's on #LMStudio today. These 7 to 12 B models are much stronger than I thought. Some of them run pretty fast, but some larger models are burning my #rtx4060 #GPU. I think I will settle with #IBM #Granite 3.3 which is a 8B model but was further trained by #unsloth to 9B. Granite 3.3 came out in April this year. In the long run, I will need a 20 to 40B model. But then I most likely need an rtx 5090 machine with 64G VRAM to run them. #AI #AIs

Train your own R1 reasoning model with Unsloth.
"We've enhanced the entire GRPO process, making it use 80% less VRAM than Hugging Face + FA2. This allows you to reproduce R1-Zero's "aha moment" on just 7GB of VRAM using Qwen2.5 (1.5B)"
#ai #reasoning #unsloth #opensource #locally #grpo
https://unsloth.ai/blog/r1-reasoning

"With 15GB VRAM, Unsloth allows you to transform any model up to 15B parameters like Llama 3.1 (8B), Phi-4 (14B), Mistral (7B) or Qwen2.5 (7B) into a reasoning model"

Train your own R1 reasoning model with Unsloth

https://unsloth.ai/blog/r1-reasoning

#LocalLLM #LLM #reasoning #unsloth #GRPO