File size: 2,874 Bytes
646b8b0
 
 
 
aeec2f0
646b8b0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
208e4eb
 
 
 
 
 
 
 
 
646b8b0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
49
50
---
license: apache-2.0
language:
  - en
base_model: tensoropera/Fox-1-1.6B
---

## Model Card for Fox-1-1.6B-Instruct

> [!IMPORTANT]
> This model is an instruction tuned model which requires alignment before it can be used in production. We will release
> the chat version soon.

Fox-1 is a decoder-only transformer-based small language model (SLM) with 1.6B total parameters developed
by [TensorOpera AI](https://tensoropera.ai/). The model was pre-trained with a 3-stage data curriculum on 3 trillion
tokens of text and code data in 8K sequence length. Fox-1 uses Grouped Query Attention (GQA) with 4 key-value heads and
16 attention heads for faster inference.

Fox-1-Instruct-v0.1 is an instruction-tuned (SFT) version of Fox-1-1.6B that has an 8K native context length. The model
was finetuned with 5B tokens of instruction following and multi-turn conversation data.

For the full details of this model please read
our [release blog post](https://blog.tensoropera.ai/tensoropera-unveils-fox-foundation-model-a-pioneering-open-source-slm-leading-the-way-against-tech-giants).

## Getting-Started

The model and a live inference endpoint are available on
the [TensorOpera AI Platform](https://tensoropera.ai/models/1228?owner=tensoropera).

For detailed deployment instructions, refer to
the [Step-by-Step Guide](https://blog.tensoropera.ai/how-to/how-to-deploy-fox-1-on-tensoropera-ai-a-step-by-step-guide-2/)
on how to deploy Fox-1-Instruct on the [TensorOpera AI Platform](https://tensoropera.ai/).

## Benchmarks

We evaluated Fox-1 on ARC Challenge (25-shot), HellaSwag (10-shot), TruthfulQA (0-shot), MMLU (5-shot),
Winogrande (5-shot), and GSM8k (5-shot). We follow the Open LLM Leaderboard's evaluation setup and report the average
score of the 6 benchmarks. The model was evaluated on a machine with 8*H100 GPUs.

|               | Fox-1-1.6B-Instruct-v0.1 | Fox-1-1.6B | Qwen1.5-1.8B-Chat | Gemma-2B-It | OpenELM-1.1B-Instruct |
|---------------|--------------------------|------------|-------------------|-------------|-----------------------|
| GSM8k         | 39.20%                   | 36.39%     | 18.20%            | 4.47%       | 0.91%                 |
| MMLU          | 44.99%                   | 43.05%     | 45.77%            | 37.70%      | 25.70%                |
| ARC Challenge | 43.60%                   | 41.21%     | 38.99%            | 43.34%      | 40.36%                |
| HellaSwag     | 63.39%                   | 62.82%     | 60.31%            | 62.72%      | 71.67%                |
| TruthfulQA    | 44.12%                   | 38.66%     | 40.57%            | 45.86%      | 45.96%                |
| Winogrande    | 62.67%                   | 60.62%     | 59.51%            | 61.33%      | 61.96%                |
| Average       | 49.66%                   | 47.13%     | 43.89%            | 42.57%      | 41.09%                |