File size: 1,964 Bytes
c3fa147
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
efda0eb
 
 
c3fa147
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
---
license: apache-2.0
base_model: FINAL-Bench/Darwin-36B-Opus
tags:
- vkae
- inference-acceleration
- serving
- moe
pipeline_tag: text-generation
---

# Darwin-36B-Opus — VKAE Accelerated

Ready-to-run, **VKAE-accelerated serving** of **Darwin-36B-Opus**, VIDRAFT's house 36B Mixture-of-Experts model. Model weights and an optimized serving runtime in a single self-contained container.

> **VKAE (VIDRAFT Kernel Acceleration Engine)** is VIDRAFT's proprietary inference-serving optimization. The acceleration recipe is withheld; only the reproducible results are published here.

## Measured performance

NVIDIA B200, single GPU, bf16, same-harness before/after.

| Metric | Baseline | VKAE | Gain |
| --- | --- | --- | --- |
| Single-stream throughput | 25.0 tok/s | **280.8 tok/s** | **11.2×** |
| Output quality | reference | preserved | no degradation |

## Quick start

```bash
docker pull vidraft/darwin36-vkae:281
docker run --gpus all -p 8000:8000 vidraft/darwin36-vkae:281
```

The container serves an **OpenAI-compatible API** on port 8000 — point any OpenAI client at `http://localhost:8000/v1`. A Blackwell (B200) or Hopper (H100/H200) class GPU is recommended.


> 📦 **Ready-to-use files in this repo:** [`Dockerfile`](./Dockerfile), [`docker-compose.yml`](./docker-compose.yml), [`run_docker.sh`](./run_docker.sh) — pull-and-run, no build required.

## Links

- **Live acceleration leaderboard** → [VIDraft/vkae](https://huggingface.co/spaces/VIDraft/vkae)
- **Docker image** → [hub.docker.com/r/vidraft/darwin36-vkae](https://hub.docker.com/r/vidraft/darwin36-vkae)
- **Collection** → [FINAL-Bench · VKAE Accelerated](https://huggingface.co/collections/FINAL-Bench/vkae-accelerated-6a47231d7e7999dd8227675a)

## About

Darwin-36B-Opus is a VIDRAFT house model (36B Mixture-of-Experts). This card documents VIDRAFT's accelerated **serving** of the model; the acceleration method is proprietary and not distributed in source form.