HF Viewer · Model Architecture Explorer
🟩
21
Interactive architecture graph for any HF model
🔄 In a Training Loop
gbhf
gbuzhf
AI & ML interests
None yet
Recent Activity
liked a Space about 3 hours ago
embedl/hfviewer new activity about 4 hours ago
peculiar-ragdoll/Cyber-Tiel-Coder-35B-A3B-GGUF-MTP:Base Tiel vs Cyber TielOrganizations
None yet
liked a
Space about 3 hours ago
Base Tiel vs Cyber Tiel
15
#3 opened 1 day ago
by 
nmh2k3
New activity in gbuzhf/Ornith-1.5-35B-A3B-Abliterated-CyberTiel-Calibrated-MTPv2-ICE-GGUF about 5 hours ago
Tiel's 27K full imatrix corpus with ICE quantization
❤️🔥 2
1
#1 opened about 5 hours ago
by gbuzhf
upvoted an article about 5 hours ago
Article
Per-tensor layout maps for GGUF quantization
bartowski
• • 14 commented on Per-tensor layout maps for GGUF quantization about 5 hours ago
Good article. That was quite the reason why I was measuring the compounding errors, and how far an error travels in MoE structure. Then decided some pareto optimal tier range, finally stopped at 19, 21, 23, 25 GB.
Others aren't really needed around 26-35B MoE range I've measured.
It's not about sizes, it's about "protection of model's original decision matrix"
Protecting the model for "correct evaluation of context, correct selection of experts, and correct routing."
The rest is evenly distributable within near quantization levels.
https://huggingface.co/gbuzhf/ICE-quantization
An example:
https://huggingface.co/gbuzhf/Ornith-1.5-35B-A3B-Abliterated-MTP-UD-APEX-GGUF
published a
model about 21 hours ago
updated a
model about 22 hours ago
card vs. json
#2 opened about 22 hours ago
by gbuzhf
Chunks...
1
#14 opened 1 day ago
by gbuzhf
liked a
model 1 day ago
upvoted 2
collections 1 day ago
*Original Version*
1
#1 opened 4 days ago
by gbuzhf
liked 2
models 3 days ago
published a
model 4 days ago
updated a
model 4 days ago
upvoted an article 9 days ago
Article
Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original
MultiverseComputingCAI
• • 50