Downloads · 30 days
195
8% of all-time downloads
OrionLLM_GRM-3.2-Sky-GGUF-el4-MTP
thread13/OrionLLM_GRM-3.2-Sky-GGUF-el4-MTP
OrionLLM_GRM-3.2-Sky-GGUF-el4-MTP is a image-text-to-text model from thread13. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product.
Recently I was running some perplexity calculations (extending initial benchmarks started here), and one thing that stood out was the quality of el4 ONYX models, especially its "quality" edition.
All-time downloads
2.3K
Public
Repo size
81 GB
Likes
1
Public
Repo makeup
Click a slice to open those files.
.gguf81 GB · 100%
At a glance
- Task
- Image-Text-to-Text
- Access
- Public
- Created
- Sep 23, 2026
- Updated
- Sep 25, 2026
- SHA
- 2a93132c
Try a prompt
Base models
Model card
From the Hugging Face model README
TL;DR
Recently I was running some perplexity calculations (extending initial benchmarks started here),
and one thing that stood out was the quality of el4 ONYX models, especially its "quality" edition.
(Another thing that stood out was somewhat unfairly ignored quality of Kwaipilot/KAT-Coder-V2.5-Dev and OrionLLM/GRM-3.2-Sky.)
Now while el4 has an xspark draft model in their repository,
I am somehow more used to the now-classic MTP one -- or may be I just somehow had more luck with mtp draft models on my specific hardware.
Either way, I used a slightly corrected version of tooltd's 'grm_mtp_graft.py to do the same trick and graft a Qwen3.6 MTP draft model by a4lg to the ONYX set made by el4, and am uploading these here.
I shall probably also upload here the modified Python code and respective perplexity measurements, which could be of general interest.
- Task
- Image-Text-to-Text
- Created
- Sep 23, 2026
- Updated
- Sep 25, 2026