73 lines
1.7 KiB
Markdown
73 lines
1.7 KiB
Markdown
---
|
|
license: apache-2.0
|
|
tags:
|
|
- Aurora
|
|
---
|
|
|
|
|
|
|
|
# Aurora-10.7b_Base
|
|
|
|
Aurora-10.7b_Base is a merge of the following models: to create a 10.7b base model that can be trained.
|
|
* [mistralai/Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)
|
|
* [mistralai/Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)
|
|
|
|
## Merged Evals: (Has Not Been Finetuned)
|
|
Aurora-10.7b_Base
|
|
* Avg: 63.98
|
|
* ARC: 62.88
|
|
* HellaSwag: 83.99
|
|
* MMLU: 60.24
|
|
* T-QA: 67.84
|
|
* Winogrande: 76.4
|
|
* GSM8K: 32.52
|
|
|
|
|
|
## (OG)Donated Evals:
|
|
Mistral-7b-v0.2
|
|
* Avg: 65.71
|
|
* ARC: 63.14
|
|
* HellaSwag: 84.88
|
|
* MMLU: 60.78
|
|
* T-QA: 68.26
|
|
* Winogrande: 77.19
|
|
* GSM8K: 40.03
|
|
|
|
## 🧩 Configuration
|
|
|
|
```
|
|
slices:
|
|
- sources:
|
|
- model: mistralai/Mistral-7B-Instruct-v0.2
|
|
layer_range: [0, 24]
|
|
- sources:
|
|
- model: mistralai/Mistral-7B-Instruct-v0.2
|
|
layer_range: [8, 32]
|
|
merge_method: passthrough
|
|
dtype: bfloat16
|
|
```
|
|
|
|
## 💻 Usage
|
|
|
|
```python
|
|
!pip install -qU transformers accelerate
|
|
|
|
from transformers import AutoTokenizer
|
|
import transformers
|
|
import torch
|
|
|
|
model = "Steelskull/Aurora_base_test"
|
|
messages = [{"role": "user", "content": "What is a large language model?"}]
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model)
|
|
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
|
pipeline = transformers.pipeline(
|
|
"text-generation",
|
|
model=model,
|
|
torch_dtype=torch.float16,
|
|
device_map="auto",
|
|
)
|
|
|
|
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
|
|
print(outputs[0]["generated_text"])
|
|
``` |