Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model:
|
| 3 |
+
- OrionLLM/OxCoder-9B
|
| 4 |
+
tags:
|
| 5 |
+
- omlx
|
| 6 |
+
- mlx
|
| 7 |
+
---
|
| 8 |
+
## about
|
| 9 |
+
oq6e quantized version of OxCoder
|
| 10 |
+
|
| 11 |
+
## imatrix used
|
| 12 |
+
Default oMLX imatrix
|
| 13 |
+
|
| 14 |
+
## rough performance metrics from quick tests/benchmarks
|
| 15 |
+
- roughly 445 tks/s prefill, 16 tk/s decode on m3 pro 36gb ram in practice on a fresh pi agent
|
| 16 |
+
- context benchmark from omlx for 131k context length started at 2k t/s prefill, finished at ~600 t/s
|
| 17 |
+
- throughput benchmark did not upload because of certain flags used but roughly matched above numbers at 128k
|
| 18 |
+
- ANE prefill with tweaked settings
|
| 19 |
+
- SpecPrefill with qwen3.5 0.8b model
|