shing3232

• MLA optimization with flashattention for llama.cpp,MLA + FA now only uses K-cache - 47% saving on KV-cache size

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 45 comments

• New DeepseekV3 as well

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 11 comments

• Deepsee launch new DSv3 as well

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 0 comments

• Fine-tuning LLMs to 1.58bit: extreme quantization experiment

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 14 comments

• Day 6: One More Thing, DeepSeek-V3/R1 Inference System Overview

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 42 comments

• Deepseek-R1 and Deepseek-R1-zero repo is preparing to launch？

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 5 comments

• Qwen2.5: A Party of Foundation Models!

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 234 comments

• New paper noise_step: TRAINING IN 1.58B WITH NO GRADIENT MEMORY

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 14 comments

• Deepseekv3 release base model

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 5 comments

• Looks like deepseekv3 API is up

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 24 comments

• Codeqwen 1.5 is out with GQA

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 13 comments

• QWEN1.5 110B just out!

  Posted by shing3232@reddit | LocalLLaMA | View on Reddit | 9 comments