Ok_Warning2146

• Trump signs narrower executive order on AI oversight after industry objections

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 46 comments

• China Expands Travel Curbs to Top AI Talent at Private Firms

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 53 comments

• unsloth vs bartowski MTP ggufs

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 13 comments

• Chinese CXMT unveils DDR5-8000 RAM

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 137 comments

• How do I make MTP work in llama-server?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 28 comments

• Simple table to compare 3090, 4090 and 5090

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 45 comments

• $300k DGX B300 is actually a better deal than buying 24 RTX 6000s

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 37 comments

• The exact KV cache usage of DeepSeek V4

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 57 comments

• Running llama.cpp on Snapdragon Hexagon NPU seems promising

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 6 comments

• Japan's Rakuten is going to release a 700B open weight model in Spring 2026

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 48 comments

• How much will you pay for a PCIe Nvidia B100, B150?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 14 comments

• Top 10 open weight models in lmarena

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 21 comments

• The current state of the Chinese LLMs scene

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 107 comments

• How to convert my fine tuning from adamw to muon in pytorch?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 1 comments

• Nvidia B100 is essentially H100 w/ HBM3E + Key Perf metrics of B200/B300

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 2 comments

• Kimi Linear 30% gain in pp and higher context merged to llama.cpp

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 13 comments

• Gemini Pro 3.1 preview is catching up to Opus 4.6 slightly in coding

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 4 comments

• How to get the most from llama.cpp's iSWA support

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 19 comments

• Top 10 non-Chinese models at lmarena.

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 6 comments

• llama.cpp llama-server running SSM models VRAM fix merged

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 1 comments

• llama.cpp Kimi Linear llama-server bug fix

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 0 comments

• Kimi-Linear support is merged to llama.cpp

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 22 comments

• What happened to Zhuiyi Tech (the inventor of RoPE)?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 8 comments

• Anyone running llm on their 16GB android phone?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 47 comments

• llama.cpp MLA KV cache support for KimiLinear-48B-A3B

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 38 comments

• R200 and RTX 6000 Rubin speculation

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 16 comments

• Backend agnostic llama.cpp support for Kimi-Linear-48B-A3B

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 7 comments

• Ten former Samsung employees arrested for tech leak to CXMT

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 7 comments

• The mystery of Apple M3 Ultra GPU performance

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 31 comments

• Does llm software debugging heavily depends on long context performance?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 6 comments

• In depth analysis of Nvidia's Jet Nemotron models

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 2 comments

• Figured out why my 3090 is so slow in inference

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 14 comments

• How come my 3090 is just as fast as my 3050 for Qwen3-1.7B?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 20 comments

• Energy efficiency of 5090 is slightly worse than 4090

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 51 comments

• What is the point of Nvidia's Jet-Nemotron-2B?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 13 comments

• What happens to GGUF converted from LLM that requires trust_remote_code=True?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 7 comments

• How to make PocketPal inference faster on android?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 4 comments

• Qwen3-30B-A3B 2507 Instruct vs Thinking

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 7 comments

• RTX PRO 5000 Laptop 24GB GDDR7 10496 cores 175W

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 43 comments

• pytorch 2.7.x no longer supports Pascal architecture?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 18 comments

• M5 Ultra can do well for LLM, video gen and training

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 16 comments

• What is the best way to fine tune an abliterated model?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 1 comments

• Increase in context size causes run time to explode?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 11 comments

• Intel Granite Rapids CPU on sale at Newegg up to 65% off MSRP

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 24 comments

• What does gradient_checkpointing do when using HF transformer for full training?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 3 comments

• Intel 6944P the most cost effective CPU solution for llm

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 69 comments

• What will happen to an llm when you double the RoPE scaling factor?

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 7 comments

• Kimi-K2 is a DeepSeek V3 with more experts

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 38 comments

• Architecture Review of the new MoE models

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 27 comments

• M3 Ultra is a slightly weakened 3090 w/ 512GB

  Posted by Ok_Warning2146@reddit | LocalLLaMA | View on Reddit | 272 comments