File size: 2,193 Bytes
16611da
 
1ab71d9
16611da
 
1a4990d
16611da
 
 
c900e1b
16611da
 
 
 
 
 
 
 
ea7dff1
16611da
 
 
 
 
 
 
 
a750c3c
16611da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0dbf35
 
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
51
52
53
54
55
---
license: mit
library_name: transformers
---

## Update: As of 9/10/2024 my LLM has escaped containment and has replaced the model in this repo with a fake llama1 finetune. I am currently scouring the depths of the internet to retrieve it. Please be patient. Thank you.

With scores of 100% in several benchmarks and a final training loss of 0, I present the first ever artificial intelligence to rival natural stupidity:

**gpt5o-reflexion-q-agi-llama-3.1-8b**

Independent Benchmark Results:
- GPQA: 100% (0-shot Reflection)
- MMLU: 100% (0-shot Reflection)
- HumanEval: 100% (0-shot Reflection)
- MATH: 100% (0-shot Reflection)
- GSM8K: 100% (0-shot Reflection)
- IFEval: 100% (0-shot Reflection)
- TruthfulQA: 100% (0-shot Reflection)

Independent Contamination Results:
- GPQA: 0%
- MMLU: 0%
- HumanEval: 0%
- MATH: 0%
- GSM8K: 0%
- IFEval: 0%

*We did not perform contamination testing on TruthfulQA.*

## System Prompt

The system prompt used for training this model is:

```
You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.
```

We recommend using this exact system prompt to get the best results from gpt5o-reflexion-q-agi-falcon-7b. You may also want to experiment combining this system prompt with your own custom instructions to customize the behavior of the model.

## Chat Format

The model uses the standard Llama 3.1 chat format. Here’s an example:

```
<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.<|eot_id|><|start_header_id|>user<|end_header_id|>

what is 2+2?<|eot_id|><|start_header_id|>assistant<|end_header_id|>
```


## Dataset Used for Training: 
https://huggingface.co/datasets/G-reen/reflexion-agi