1. Home
  2. Detail

@tahanoutdoor: Bawa bekalan air, jangan lupa dispenser sekali! 💦🎒 TAHAN Automatic Water Dispenser memang sesuai untuk camping, picnic dan aktiviti outdoor supaya lebih mudah nak ambil air bila dahaga. 🏕️✨

TAHAN outdoor
TAHAN outdoor
Open In TikTok:
Region: MY
Tuesday 13 October 2026 02:00:00 GMT
286
3
0
0

Music

Download

No Watermark .mp4 (8.67MB) No Watermark(HD) .mp4 (4.82MB) Watermark .mp4 (8.97MB) Music .mp3

Comments

There are no more comments for this video.
To see more videos from user @tahanoutdoor, please go to the Tikwm homepage.

Other Videos

Bộ nhông xích #nguyenhoaixinhtuoi
Bộ nhông xích #nguyenhoaixinhtuoi
Lee Kuan Yew - let's have a sense of proportion. Let's live and let live #sg #leadership #singapore #chinese #indian
Lee Kuan Yew - let's have a sense of proportion. Let's live and let live #sg #leadership #singapore #chinese #indian
#creatorsearchinsights #فيهذااليوم
#creatorsearchinsights #فيهذااليوم
হে আমার রব আমার জন্য এমন ফয়সালা করুন...!
হে আমার রব আমার জন্য এমন ফয়সালা করুন...!
🦀🦀🦀
🦀🦀🦀
AirLLM runs 70B on a 4GB GPU Repo: github.com/lyogavin/airllm A 70B model usually needs a GPU you don't have. AirLLM streams the model's layers off disk, one at a time, so it runs on a single 4GB card instead, no quantization, no distillation, no pruning. One line of code loads Qwen, DeepSeek, Llama, and more. The same trick trains a 125B model under 6GB, and DeepSeek V3 (671B parameters) runs in about 12GB. Kimi K3, the largest open-source model yet, fits under 4GB. It takes specific builds to hit those numbers: Kimi K3 needs flash-attn and a CUDA 12 build, and the largest configs need transformers installed from source plus 360GB of free disk space. This and a lot more on our Telegram, link in profile. Telegram: t.me/SkipCI #AirLLM #OpenSource #LLM #GPU #DeepSeek #Llama #Qwen #MachineLearning
AirLLM runs 70B on a 4GB GPU Repo: github.com/lyogavin/airllm A 70B model usually needs a GPU you don't have. AirLLM streams the model's layers off disk, one at a time, so it runs on a single 4GB card instead, no quantization, no distillation, no pruning. One line of code loads Qwen, DeepSeek, Llama, and more. The same trick trains a 125B model under 6GB, and DeepSeek V3 (671B parameters) runs in about 12GB. Kimi K3, the largest open-source model yet, fits under 4GB. It takes specific builds to hit those numbers: Kimi K3 needs flash-attn and a CUDA 12 build, and the largest configs need transformers installed from source plus 360GB of free disk space. This and a lot more on our Telegram, link in profile. Telegram: t.me/SkipCI #AirLLM #OpenSource #LLM #GPU #DeepSeek #Llama #Qwen #MachineLearning

About

Legal