Laxmi Kant
laxmimerit
·
AI & ML interests
None yet
Organizations
Qwen-Image-2.1: Quantized Inference Measurements on a Single RTX 5090
🚀🔥 2
#32 opened 2 months ago
by
laxmimerit
Reading GLM 5.3 Flash from config.json: 6.0 KB of KV Cache per Token
1
#55 opened 2 months ago
by
laxmimerit
Reading DeepSeek V4.1 Flash's config.json: Where Sparse Attention Lives
❤️🔥 1
#38 opened 2 months ago
by
laxmimerit
I Benchmarked on RTX 5090: 271 tok/s generation, 673 tok/s at 8 concurrent, 32 KB/token KV cache
2
#31 opened 3 months ago
by
laxmimerit
Run Qwen 3.8 Flash Next (Qwen 4) 180B on CPU Only, No GPU Needed
🔥❤️ 9
3
#29 opened 3 months ago
by
laxmimerit
Qwen3.8-Flash-Next (Qwen 4 - Preview) vs 3.8 27B - Architecture Teardown
🔥❤️ 8
6
#22 opened 3 months ago
by
laxmimerit
Benchmarked Qwen3.8-27B vs Nemotron 3.5 Lightning and Muse Glimmer on 16 hard problems (RTX 5090, Q4_K_M)
🔥❤️ 11
13
#103 opened 3 months ago
by
laxmimerit
This model cannot stop thinking
🧠➕ 9
6
#113 opened 3 months ago
by
xvcy3w
Follow-up: I benchmarked 45 llama.cpp settings on Qwen3.8-27B. The MTP draft head is worth 1.81x, free (RTX 5090, Q4_K_M)
🚀❤️ 9
2
#112 opened 3 months ago
by
laxmimerit