Downloads · 30 days
80
11% of all-time downloads
greenpolo/langslice-gemma-4-E4B
langslice-gemma-4-E4B is a image-text-to-text model from greenpolo. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as gemma.
Downloads · 30 days
80
11% of all-time downloads
All-time downloads
715
Public
Parameters
8B
45.4 GB on disk
Likes
0
Public
Click a slice to open those files.
.gguf29.4 GB · 65%
How the weights are stored.
BF168B · 100%
From the Hugging Face model README

A Gemma 4 E4B-IT fine-tune that estimates the anterior-posterior (AP) coordinate of a histological brain section in the matching mouse coronal atlas. Trained on adult (Allen mouse 25 µm) and developmental (DeMBA P9–P35, ADMBA P4–P56) coronal sections.
| Model | Mean abs. error (mm) | Mean error (% of 13.2 mm AP axis) |
|---|---|---|
| Base unsloth/gemma-4-E4B-it | 2.375 | 18.02 % |
| LangSlice-Gemma-4-E4B v1.0 | 1.228 | 9.32 % |
| Gemini 3 Flash Preview | 0.429 | 3.25 % |
SliceBench small coronal, mouse subset (n = 112).
vllm serve greenpolo/langslice-gemma-4-E4B \
--max-model-len 16384 \
--enable-auto-tool-choice \
--tool-call-parser gemma4 \
--reasoning-parser gemma4
GGUF (Q4_K_M, Q8_0, BF16) under gguf/ for use with
llama.cpp llama-mtmd-cli (pair with
the bundled mmproj-BF16.gguf for image input).
The model is intentionally scoped to single-slice mouse coronal AP estimation. Sagittal, horizontal, and rat performance are much weaker. Inference must use the bundled chat template.
Full agent loop, multi-slice estimation, and end-to-end registration: LangSlice harness.
Gemma 4 Terms of Use governs the weights.
A Zenodo DOI is forthcoming. Until it's minted, please cite this Hugging Face repository directly:
Baughman, N. and Walsh Lab. LangSlice-Gemma-4-E4B v1.0. Hugging Face, 2026. https://huggingface.co/greenpolo/langslice-gemma-4-E4B