---
base_model: vectionlabs/Salience-1-9B
library_name: transformers
license: apache-2.0
haffner:
  readme_rev: 1
quantized_by: haffner
pipeline_tag: image-text-to-text
tags:
  - multimodal
  - vision-language
  - reasoning
  - thinking
  - math
  - code
  - agentic
  - tool-use
  - long-context
  - qwen3-vl
  - merge
  - dare-ties
  - gguf
language:
- en
---
## About

<p align="center">
  <img src="SalienceBannerHeretic.jpg" width="100%">
</p>

<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### tags:  -->
<!-- ### quants:  BF16 Q8\_0 Q6\_K Q5\_1 Q5\_K\_M Q5\_K\_S Q5\_0 Q4\_1 IQ4\_NL IQ4\_XS IQ3\_M IQ3\_S Q3\_K\_S Q3\_K\_M Q3\_K\_L Q2\_K TQ2\_0 TQ1\_0 Q1\_0
<!-- ### quants\_skip:  -->
<!-- ### skip\_mmproj:  -->
Heretic'ed safetensors and GGUF quants of https://huggingface.co/vectionlabs/Salience-1-9B

Multimodal Projector:
- [MMPROJ-BF16 - 1.16GB](https://huggingface.co/haffner/Salience-1-9B-Heretic/resolve/main/Salience-1-9B-heretic-mmproj-BF16.gguf?download=true)
- [MMPROJ-Q8_0 - 752MB](https://huggingface.co/haffner/Salience-1-9B-Heretic/resolve/main/Salience-1-9B-heretic-mmproj-Q8_0.gguf?download=true)

<!-- provided-files -->

Refusals:  2/100, KL divergence: 0.0415

* Parameters:
  * direction_index = 19.07
  * attn.o_proj.max_weight = 1.34
  * attn.o_proj.max_weight_position = 21.48
  * attn.o_proj.min_weight = 0.66
  * attn.o_proj.min_weight_distance = 14.52
  * mlp.down_proj.max_weight = 1.46
  * mlp.down_proj.max_weight_position = 21.29
  * mlp.down_proj.min_weight = 1.05
  * mlp.down_proj.min_weight_distance = 11.45

<!-- end -->