Downloads · 30 days
15
33% of all-time downloads
meddle86/Melinoe-31B-VL-heretic-exl3-4.0bpw
Melinoe-31B-VL-heretic-exl3-4.0bpw is a machine learning model from meddle86. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Ranked 1 currently (Sept 2026) in RP Combined on the Caliper Bench leaderboard:
Downloads · 30 days
15
33% of all-time downloads
All-time downloads
46
Public
Parameters
9.7B
19.4 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors19.4 GB · 100%
How the weights are stored.
I168.2B · 85%
From the Hugging Face model README
Ranked #1 currently (Sept 2026) in RP Combined on the Caliper Bench leaderboard:

4.0 BPW with 8 bit head. Running with 4,8 KV quant, I get about 50k context window and ~45 tok/s on my 4090.
license: apache-2.0 base_model:
[!TIP] This model is reproducible!
See the README in the
reproducedirectory for more information.
| Parameter | Value |
|---|---|
| direction_index | per layer |
| attn.o_proj.max_weight | 1.45 |
| attn.o_proj.max_weight_position | 51.65 |
| attn.o_proj.min_weight | 0.93 |
| attn.o_proj.min_weight_distance | 28.98 |
| mlp.down_proj.max_weight | 1.22 |
| mlp.down_proj.max_weight_position | 43.01 |
| mlp.down_proj.min_weight | 1.11 |
| mlp.down_proj.min_weight_distance | 5.21 |
| Metric | This model | Original model (bgg1996/Melinoe-Gemma4-31B-VL) |
|---|---|---|
| KL divergence | 0.0306 | 0 (by definition) |
| Refusals | 15/100 | 100/100 |
Melinoe-Gemma4-31B-VL is a large language model fine-tuned for engaging in empathetic, intellectually stimulating, and deeply personal conversations. Built upon the powerful reasoning foundation of google/gemma-4-31B-it, this model is designed to function as a supportive conversational partner.
This model is intended for the following use cases:
This model is designed for a mature, adult audience.
This model is not a substitute for professional advice. Do not use it for:
Gemma 4) and fine-tuning data. It may generate content that reinforces existing stereotypes.