Ben Kelly's picture

In a Training Loop 🔄

Ben Kelly PRO

YellowjacketGames

·

manacasterben

AI & ML interests

None yet

Recent Activity

replied to danielhanchen's post about 7 hours ago

You can now run Kimi K2.5 locally! 🔥 We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit. Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision. GGUF: https://huggingface.co/unsloth/Kimi-K2.5-GGUF Guide: https://unsloth.ai/docs/models/kimi-k2.5

replied to danielhanchen's post about 8 hours ago

You can now run Kimi K2.5 locally! 🔥 We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit. Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision. GGUF: https://huggingface.co/unsloth/Kimi-K2.5-GGUF Guide: https://unsloth.ai/docs/models/kimi-k2.5

updated a collection about 9 hours ago

[mixed] ORCAssist "Work's Done!"

View all activity

Organizations

replied to danielhanchen's post about 7 hours ago

people commenting about how they can't run this locally are saying more about themselves than they are about anything else

wheres the spirit of innovation? get out a PC from 1997, try to get it running LM Studio, never know if you dont try!

replied to danielhanchen's post about 8 hours ago

I'll give this a try on our office GPU rig. VRAM limited to 96gb but DRAM is 1024gb attached to a 64 core threadripper.

I doubt I'll be able to pull 40 TPS on this rig, but hey, local is local!

Thx to all the Unsloth guys as usual!

updated a collection about 9 hours ago

[mixed] ORCAssist "Work's Done!"

16 items • Updated about 9 hours ago • 1

upvoted a paper about 9 hours ago

ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration

Paper • 2511.21689 • Published Nov 26, 2025 • 121

upvoted 5 articles about 20 hours ago

Article

MCP for Research: How to Connect AI to Research Tools

Aug 18, 2025

•

66

Article

VibeGame: Exploring Vibe Coding Games

Sep 29, 2025

•

37

Article

Building the Open Agent Ecosystem Together: Introducing OpenEnv

+8

Oct 23, 2025

•

148

Article

DeepMath: A lightweight math reasoning Agent with smolagents

+1

Dec 4, 2025

•

37

Article

Diffusers welcomes FLUX-2

+6

Nov 25, 2025

•

176

upvoted an article about 21 hours ago

Article

The Open Evaluation Standard: Benchmarking NVIDIA Nemotron 3 Nano with NeMo Evaluator

Dec 17, 2025

•

46

updated a collection about 21 hours ago

[mixed] ORCAssist "Work's Done!"

16 items • Updated about 9 hours ago • 1

upvoted a paper about 21 hours ago

WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents

Paper • 2509.06501 • Published Sep 8, 2025 • 80

updated a collection about 21 hours ago

[models] CPU-Offload &/|| A6000x2

TPS can be as low as 1.0, seriously. its SLOW. • 6 items • Updated about 21 hours ago • 1

updated a collection 1 day ago

[papers] Film & Cinema

2 items • Updated about 21 hours ago

upvoted a paper 1 day ago

Advancing Open-source World Models

Paper • 2601.20540 • Published 2 days ago • 85

updated a collection 1 day ago

[papers] Gameplay Optimization

Research papers that may contribute to a broader approach to teaching machines how to play complex strategy games beyond just Chess. • 16 items • Updated about 21 hours ago • 1

upvoted a paper 1 day ago

World Craft: Agentic Framework to Create Visualizable Worlds via Text

Paper • 2601.09150 • Published 17 days ago • 19

updated a collection 1 day ago

[papers] Distillation

6 items • Updated about 21 hours ago • 1

upvoted a paper 1 day ago

Reinforcement Learning via Self-Distillation

Paper • 2601.20802 • Published 2 days ago • 20

updated a collection 1 day ago

[papers] Distillation

6 items • Updated about 21 hours ago • 1