d3 Core ML: README (vllm-sr Decision 3.0 v3.1.0)
Browse files
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- vllm-sr/d3-lite
|
| 5 |
+
- vllm-sr/d3-nano
|
| 6 |
+
- vllm-sr/d3-mini
|
| 7 |
+
library_name: coreml
|
| 8 |
+
pipeline_tag: zero-shot-classification
|
| 9 |
+
tags:
|
| 10 |
+
- coreml
|
| 11 |
+
- apple-silicon
|
| 12 |
+
- decision-model
|
| 13 |
+
- vision
|
| 14 |
+
- video
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# Decision 3.0 — Core ML
|
| 18 |
+
|
| 19 |
+
[Decision 3.0](https://huggingface.co/collections/vllm-sr/decision-30) by the vLLM Semantic Router team, converted to Core ML for Apple silicon: text, image and video requests on device. Same contract as upstream: one forward pass per question, the 255-way FP32 readout at the last prompt token, a probability for every option, no text generated.
|
| 20 |
+
|
| 21 |
+
| Folder | Upstream | Size |
|
| 22 |
+
|---|---|---|
|
| 23 |
+
| [`lite/`](lite) | [vllm-sr/d3-lite](https://huggingface.co/vllm-sr/d3-lite) | 0.8B |
|
| 24 |
+
| [`nano/`](nano) | [vllm-sr/d3-nano](https://huggingface.co/vllm-sr/d3-nano) | 2B |
|
| 25 |
+
| [`mini/`](mini) | [vllm-sr/d3-mini](https://huggingface.co/vllm-sr/d3-mini) | 4B |
|
| 26 |
+
|
| 27 |
+
Each folder is a full release (text graphs, vision tower, host tables, Python host) with its own README covering files, parity against upstream and speed. Swift: `D3VisionManager` in [FluidUse](https://github.com/FluidInference/FluidUse) loads any of them.
|
| 28 |
+
|
| 29 |
+
License: Apache-2.0, as upstream.
|