File size: 2,192 Bytes
8c155e1
 
 
 
 
 
 
 
69285bb
 
8c155e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: other
license_name: tongyi-qianwen
license_link: https://huggingface.co/Qwen/Qwen2-72B-Instruct/blob/main/LICENSE
language:
- en
- zh
pipeline_tag: text-generation
base_model: anthracite-org/magnum-v1-32b
quantized_by: anthracite-org
tags:
- chat
---
![image/png](https://cdn-uploads.huggingface.co/production/uploads/635567189c72a7e742f1419c/PK7xRSd18Du0bX-w_t-9c.png)
## This repo contains GGUF quants of the model. If you need the original weights, please find them [here](https://huggingface.co/anthracite-org/magnum-32b-v1).
This is the second in a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus. This model is fine-tuned on top of [Qwen1.5 32B](https://huggingface.co/Qwen/Qwen1.5-32B).


## Prompting
Model has been Instruct tuned with the ChatML formatting. A typical input would look like this:

```py
"""<|im_start|>user
Hi there!<|im_end|>
<|im_start|>assistant
Nice to meet you!<|im_end|>
<|im_start|>user
Can I ask a question?<|im_end|>
<|im_start|>assistant
"""
```

## Credits

Three new general purpose instruction following datasets were added on top of the original Stheno dataset (which had certain low quality entries purged/removed).
The first two were designed specifically for the Magnum series, to better address prompt adherence and coherence:
- [kalomaze/Opus_Instruct_25k](https://huggingface.co/datasets/kalomaze/Opus_Instruct_25k)
- [Nopm/Opus_WritingStruct](https://huggingface.co/datasets/Nopm/Opus_WritingStruct)
- [Gryphe/Sonnet3.5-SlimOrcaDedupCleaned](https://huggingface.co/datasets/Gryphe/Sonnet3.5-SlimOrcaDedupCleaned) (A ~16k rows subset)

This model has been a team effort, and the credits goes to all members of Anthracite.

## Training
The training was done for 2 epochs with a learning rate of 1e-05. We used 8x [NVIDIA H100 Tensor Core](https://www.nvidia.com/en-us/data-center/h100/) GPUs for the full-parameter fine-tuning of the model.

[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)

## Safety
...