👋 Open to Work
Kaloyan Nikolov
sleepyeldrazi
AI & ML interests
ML engineer working on LLM inference and training. Finalizing my M.Sc. Computer Science @ RWTH Aachen.
Currently focusing on DGX Spark models/frameworks. Contact: konik98@gmail.com
Recent Activity
new activity about 5 hours ago
ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF:Are the ngrams quanted to low quants as well or does it have q8 ngrams? new activity about 5 hours ago
ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF:Great quantization method, but model is too wild to use anywhereOrganizations
None yet
Are the ngrams quanted to low quants as well or does it have q8 ngrams?
👍 4
1
#34 opened 2 days ago
by 
clover-supply
Great quantization method, but model is too wild to use anywhere
1
#37 opened about 7 hours ago
by 
krustik
Does this increase the quality too or just changes the data format?
1
#1 opened 3 days ago
by 
GanjaSmoker
higher precision PLE weights?
👍 3
4
#28 opened 5 days ago
by 
aba-aba
List of pruned experts
5
#1 opened about 1 month ago
by 
kdrapel
Early report running it on a DGX Spark, 4 concurrent @ 256k bf16 with SSD offload
👍 7
4
#31 opened about 1 month ago
by sleepyeldrazi
BF16 Missing currently
1
#1 opened about 1 month ago
by sleepyeldrazi
NVME vs Optane?
3
#29 opened about 1 month ago
by 
AndyFaceHugger
Is it possible to offload n-gram to an NVMe SSD?
👀🚀 31
15
#11 opened about 1 month ago
by 
lingyezhixing
Yet to see any gains from DFlash
👍 11
13
#3 opened about 1 month ago
by 
Dnonmi
BenchMAXXXED
➕ 2
33
#28 opened 2 months ago
by 
TawusGames
Exploring ThinkingCap fine-tunes for Qwen 3.8: Is it still necessary with native `reasoning_effort`?
👍 4
2
#106 opened about 2 months ago
by 
icognito
How can I run this well on an RTX 5070 ti?
8
#36 opened about 2 months ago
by 
theonerm2
Text-only? Plans for multimodal release?
👀😔 5
2
#7 opened about 2 months ago
by sleepyeldrazi
Reproducible Laguna S 2.1 NVFP4 runaway reasoning: output budget scales 2K → 16K with 12/12 length exhaustion and no final answer
👀👍 8
7
#16 opened 2 months ago
by 
darkmatter2222
Still looping after update - details inside
👀 2
22
#10 opened 2 months ago
by 
cbert33
Model is decent, its faster than Qwen and pretty close on real world tool calls
❤️👀 3
7
#22 opened 2 months ago
by 
Jakaryun
MOE or changing to A3B
3
#43 opened 2 months ago
by 
Tom-Neverwinter