Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- bonfyre-fpq
|
| 4 |
+
- v12
|
| 5 |
+
- quantized
|
| 6 |
+
- e8-lattice
|
| 7 |
+
- rans-entropy
|
| 8 |
+
license: apache-2.0
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# DPT Large — BonfyreFPQ v12
|
| 12 |
+
|
| 13 |
+
Native `.fpq` v12 quantized weights for [Intel/dpt-large](https://huggingface.co/Intel/dpt-large).
|
| 14 |
+
|
| 15 |
+
## Model Info
|
| 16 |
+
|
| 17 |
+
| Property | Value |
|
| 18 |
+
|----------|-------|
|
| 19 |
+
| Architecture | Monocular Depth Estimation |
|
| 20 |
+
| Parameters | 344M |
|
| 21 |
+
| Original Size | 1304 MB (safetensors) |
|
| 22 |
+
| FPQ v12 Size | 329 MB |
|
| 23 |
+
| Compression | 4.0× |
|
| 24 |
+
| Avg Cosine | 0.999742 |
|
| 25 |
+
| Worst Cosine | 0.998460 |
|
| 26 |
+
|
| 27 |
+
## Format
|
| 28 |
+
|
| 29 |
+
BonfyreFPQ v12 native format: rANS entropy-coded E8 lattice coordinates with 6-bit packed tiles and FP16 scales.
|
| 30 |
+
|
| 31 |
+
### Decode to safetensors
|
| 32 |
+
|
| 33 |
+
```bash
|
| 34 |
+
bonfyre-fpqx decode dpt-large-v12-fpq.fpq output.safetensors
|
| 35 |
+
```
|
| 36 |
+
|
| 37 |
+
## Source
|
| 38 |
+
|
| 39 |
+
Converted from [Intel/dpt-large](https://huggingface.co/Intel/dpt-large) using BonfyreFPQ v12.
|