JonnaMat commited on
Commit
34aa77b
·
verified ·
1 Parent(s): 73581d3

Simplify README.md

Browse files
Files changed (1) hide show
  1. README.md +17 -40
README.md CHANGED
@@ -12,48 +12,25 @@ short_description: Embedl - efficient AI for the edge
12
 
13
  # Embedl
14
 
15
- <img src="https://huggingface.co/datasets/embedl/documentation-images/resolve/main/organization_banner.png" alt="Embedl Organization Banner" width="100%">
16
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
 
18
  Embedl develops advanced tools and algorithms for **Edge AI**. Our mission is to make AI models run
19
  **faster**, **more energy-efficient**, and **reliably across diverse hardware platforms**, while
20
  significantly reducing development time.
21
 
22
- We help teams deploy high-performance AI on real-world, resource-constrained devices.
23
-
24
-
25
- ### **Embedl Models** ([Community](https://github.com/embedl/embedl-models))
26
-
27
- Pre-optimized models that can be used **off-the-shelf** or customized for specific hardware target
28
- supported by the [embedl-models](https://github.com/embedl/embedl-models) package.
29
-
30
- **First release highlights:**
31
-
32
- - The **fastest Small Language Models (SLMs)** using **[FlashHead](https://www.embedl.com/knowledge/ultra-efficient-llms-embedls-breakthrough-for-on-device-ai)**,
33
- a novel architectural improvement to the language-model head
34
- - Works with popular models like **Llama, Gemma, and Qwen**
35
- - Provides speedups on top of:
36
- - Quantization
37
- - Flash Attention
38
- - Other standard optimizations
39
-
40
- Device: Nvidia Jetson Thor
41
- | Model | Generation speed (tokens/s) |
42
- | ------------------------------------------------ | ----------------------------|
43
- | embedl/Llama-3.2-3B-Instruct-FlashHead-W4A16 | 100 |
44
- | Llama-3.2-3B-Instruct-W4A16* | 80 |
45
- | RedHatAI/Llama-3.2-3B-Instruct-FP8 | 64 |
46
- | meta-llama/Llama-3.2-3B-Instruct | 37 |
47
-
48
- *Embedl quantized model for benchmarking similar to the FlashHead-W4A16 but without
49
- the faster FlashHead and custom generation loop.
50
-
51
- ---
52
-
53
- ## Contact
54
-
55
- **Headquarters (Sweden)**
56
- Gamla Almedalsvägen 39
57
- 412 63 Gothenburg, Sweden
58
-
59
- **Email:** contact@embedl.com
 
12
 
13
  # Embedl
14
 
15
+
16
+
17
+ <img src="https://huggingface.co/datasets/embedl/documentation-images/resolve/main/organization_banner.png" alt="Embedl Organization Banner" width="100%">
18
+
19
+ <p align="center">
20
+ <b>Efficient AI for the edge.</b>
21
+ </p>
22
+
23
+ <p align="center">
24
+ <a href="https://embedl.com"><img alt="Website" src="https://img.shields.io/badge/embedl.com-website-blue" /></a>
25
+ <a href="https://github.com/embedl"><img alt="GitHub" src="https://img.shields.io/badge/GitHub-embedl-black?logo=github" /></a>
26
+ <a href="https://arxiv.org/abs/2603.14591"><img alt="arXiv"
27
+ src="https://img.shields.io/badge/arXiv-2603.14591-b31b1b.svg?logo=arxiv" /></a>
28
+ <a href="mailto:models@embedl.com"><img alt="Contact" src="https://img.shields.io/badge/Contact-models%40embedl.com-green" /></a>
29
+ </p>
30
+
31
 
32
  Embedl develops advanced tools and algorithms for **Edge AI**. Our mission is to make AI models run
33
  **faster**, **more energy-efficient**, and **reliably across diverse hardware platforms**, while
34
  significantly reducing development time.
35
 
36
+ We help teams deploy high-performance AI on real-world, resource-constrained devices.