Card: feature-extraction pipeline, discovery tags, collection link
Browse files
README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
---
|
| 2 |
base_model: google/embeddinggemma-2
|
| 3 |
library_name: mlx
|
| 4 |
-
pipeline_tag:
|
| 5 |
license: apache-2.0
|
| 6 |
tags:
|
| 7 |
- mlx
|
|
@@ -10,12 +10,21 @@ tags:
|
|
| 10 |
- quantized
|
| 11 |
- krill
|
| 12 |
- 8bit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
---
|
| 14 |
|
| 15 |
# EmbeddingGemma 2, MLX 8bit
|
| 16 |
|
| 17 |
A quantized MLX build of [google/embeddinggemma-2](https://huggingface.co/google/embeddinggemma-2) (740M parameters; text, image, audio and video embeddings in one `model.safetensors`), made with [Krill](https://github.com/srvsngh99/Krill). Bits: **8**. Weights file: **806.2 MB** (bf16 original: 1488.9 MB).
|
| 18 |
|
|
|
|
|
|
|
| 19 |
## What this build is
|
| 20 |
|
| 21 |
Affine 8-bit quantization, group size 64, applied to every quantizable tensor.
|
|
|
|
| 1 |
---
|
| 2 |
base_model: google/embeddinggemma-2
|
| 3 |
library_name: mlx
|
| 4 |
+
pipeline_tag: feature-extraction
|
| 5 |
license: apache-2.0
|
| 6 |
tags:
|
| 7 |
- mlx
|
|
|
|
| 10 |
- quantized
|
| 11 |
- krill
|
| 12 |
- 8bit
|
| 13 |
+
- embedding
|
| 14 |
+
- multimodal-embedding
|
| 15 |
+
- multimodal
|
| 16 |
+
- vision
|
| 17 |
+
- audio
|
| 18 |
+
- video
|
| 19 |
+
- apple-silicon
|
| 20 |
---
|
| 21 |
|
| 22 |
# EmbeddingGemma 2, MLX 8bit
|
| 23 |
|
| 24 |
A quantized MLX build of [google/embeddinggemma-2](https://huggingface.co/google/embeddinggemma-2) (740M parameters; text, image, audio and video embeddings in one `model.safetensors`), made with [Krill](https://github.com/srvsngh99/Krill). Bits: **8**. Weights file: **806.2 MB** (bf16 original: 1488.9 MB).
|
| 25 |
|
| 26 |
+
All 9 builds, next to Google's original, are in the [EmbeddingGemma 2 MLX collection](https://huggingface.co/collections/srv-sngh/embeddinggemma-2-mlx-quantization-ladder-6ac7b2592bbdabdd74dc5ba7).
|
| 27 |
+
|
| 28 |
## What this build is
|
| 29 |
|
| 30 |
Affine 8-bit quantization, group size 64, applied to every quantizable tensor.
|