SL-AI/GRaPE-2.5-Helios-GGUF
Image-Text-to-Text • 9B • Updated • 241 • 2
Koitenshin
Koitenshin
AI & ML interests
I like breaking things and playing the mad scientist. Did you know we still haven't completely untapped the full potential of SD 1.5 yet?
Recent Activity
new activity 7 days ago
XHToken/Spark-X2.6-4B-GGUF:Nanobench benchmark results for Spark-X2.6-4B-GGUF new activity 7 days ago
OpenMOSS-Team/MOSS-Image-SFT:SFT vertical colour stripes: where they come from, plus a small fix new activity 29 days ago
ewald1976/G4-Plainsong-12B:Terse training works against Multi-Char RPOrganizations
None yet
Nanobench benchmark results for Spark-X2.6-4B-GGUF
🚀 4
2
#8 opened 13 days ago
by 
userr99
SFT vertical colour stripes: where they come from, plus a small fix
🧠👍 16
1
#8 opened 24 days ago
by 
nynxz
Terse training works against Multi-Char RP
👍 1
1
#1 opened 29 days ago
by Koitenshin
Tool calls and reasoning not reliable on llama.cpp (fix provided)
👍❤️ 18
8
#6 opened about 2 months ago
by 
coder543
liked a
model about 2 months ago
This model is insane.
👍 1
1
#1 opened about 2 months ago
by Koitenshin
Gemma 4.2 26B A4B?
2
#58 opened 3 months ago
by 
kozusenok
Adding `safetensors` variant of this model
🔥 4
2
#1 opened over 1 year ago
by 
SFconvertbot
Prescence penalty 1.5
👀👍 4
2
#5 opened 7 months ago
by 
Sliderpro93
The high-quality is actually pretty bad
11
#3 opened 3 months ago
by 
I-Am-Sam
liked a
model 2 months ago
Extremely slow, but amazing
#2 opened 2 months ago
by Koitenshin
not particularly smart
👍 15
1
#7 opened 5 months ago
by 
toriset
Thanks - it works well in opencode
🤝❤️ 7
13
#1 opened 5 months ago
by 
Milor123
The model suffers from ADHD
👍 1
7
#18 opened 11 months ago
by 
Candala
What about uncensored/abliterated version?
❤️ 2
12
#2 opened 6 months ago
by 
tima2431
Image Dimension Test Results
1
#27 opened 7 months ago
by Koitenshin
updated a
model about 1 year ago
commented on Migrating the Hub from Git LFS to Xet about 1 year ago
btw - is the deduplication done after upload then, or xet somehow compares diffs vs 20+ PB of data prior to upload?
Great question - some of that is covered in a section of another blog post. TL;DR, it's handled before upload using a few heuristics. Also important to note that for performance reasons, we treat deduplication as an optimization, not the foundational goal of the system. That way we can make a best effort to provide global dedupe without significantly impacting the speed of file transfers.
Which makes it absolute garbage due to the 100 GB Limit. If you hit the limit during upload to your private repo, trying to reupload the file later will fail due to invalid shards.