Aj | LLM-salad

aj9o9

AI & ML interests

Obsessed with LLMs and ways to improve them! Super obsessed with with quants now, making them more efficient. https://x.com/ItsmeAjayKV

Recent Activity

new activity 1 day ago

aj9o9/Qwen3.8-27B-DSpine-b16-GGUF:Intel Arc Pro B65 (SYCL): measured numbers - n-max 15 is best for code/tool calls (5.5-5.9x), 3 for prose

new activity 2 days ago

aj9o9/Qwen3.8-27B-DSpine-b16-GGUF:Results on AMD (Radeon AI PRO R9700, ROCm/HIP): it works, code about 2x faster, images and prose not

updated a model 2 days ago

aj9o9/Qwen3.8-27B-DSpine-b16-GGUF

View all activity

Organizations

None yet

New activity in aj9o9/Qwen3.8-27B-DSpine-b16-GGUF 1 day ago

Intel Arc Pro B65 (SYCL): measured numbers - n-max 15 is best for code/tool calls (5.5-5.9x), 3 for prose

#4 opened 2 days ago by

ashokkumards

New activity in aj9o9/Qwen3.8-27B-DSpine-b16-GGUF 2 days ago

Results on AMD (Radeon AI PRO R9700, ROCm/HIP): it works, code about 2x faster, images and prose not

#3 opened 2 days ago by

menace1000

DGX Spark (GB10) benchmarks: acceptance matches the 3090, but the drafter ranking flips on prose

#2 opened 2 days ago by

Alpaczyk

RDNA3 (RX 7800 XT / RADV Vulkan): DSpine throughput findings — spec-draft-n-max 2 is the sweet spot (1.53x)

#1 opened 2 days ago by

howdya

New activity in aj9o9/Qwen3.8-27B-Escha-W2-GGUF about 1 month ago

Errors while trying to load it with the custom llama.cpp

#3 opened about 1 month ago by

thegreatmammut

New activity in aj9o9/Qwen3.8-27B-Escha-W2-GGUF 2 months ago

Comared to Unsloth Dynamic3.0 quants

#1 opened 2 months ago by

rbtasd

kvarn possible

🧠 1

#2 opened 2 months ago by

hugless

New activity in aj9o9/GLM-5.3-Flash-GGUF 2 months ago

Where are shards

#1 opened 2 months ago by

ahmadriaz9999

New activity in aj9o9/Ling-3.0-flash-GGUF 3 months ago

Well done fella

❤️ 2

#1 opened 3 months ago by

pgib2003