Cerebras-GPT-2.7B-Alpaca-SP/README.md

---
license: apache-2.0
tags:
- code
---

## Cerebras-GPT-2.7B-Alpaca-SP

Cerebras-GPT-2.7B-Alpaca-SP is [cerebras/Cerebras-GPT-2.7B](https://huggingface.co/cerebras/Cerebras-GPT-2.7B) finetuned on a modified [Alpaca dataset](https://github.com/tloen/alpaca-lora/blob/main/alpaca_data_cleaned.json) with a shorter prompt structure (`Human:...\n\nAssistant:`). This contains the float16 model merged with [lxe/lora-cerebras-gpt2.7b-alpaca-shortprompt](https://huggingface.co/lxe/lora-cerebras-gpt2.7b-alpaca-shortprompt) LoRA.


### Limitations:

Although mostly coherent, the model tends to hallucinate signinficantly, often producing erroneous information.

### Basic Usage:

```python
tokenizer = transformers.AutoTokenizer.from_pretrained('lxe/Cerebras-GPT-2.7B-Alpaca-SP')

model = transformers.AutoModelForCausalLM.from_pretrained(
    `lxe/Cerebras-GPT-2.7B-Alpaca-SP`,
    load_in_8bit=True,
    torch_dtype=torch.float16,
    device_map={"": 0},
)

prompt = "Human: how old is the sun?\n\nAssistant:"
input_ids = tokenizer.encode(prompt, return_tensors="pt").cuda()

with torch.no_grad():
    output = model.generate(
        input_ids=input_ids,
        max_length=100,
        early_stopping=True,
    )
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)
``` 

```
Human: how old is the sun?

Assistant: The sun is approximately 4.5 billion years old.
```

### Example Outputs:

**Human**: how old is the sun? <br />
**Assistant**: The sun is approximately 4.5 billion years old.


**Human**: what are the highest grossing films of the 2000-2010 decade? <br />
**Assistant**: The highest grossing film in 2010 was Avatar, which earned $2.5 billion worldwide and is currently ranked as one of the top 10 movies ever made by Box Office Mojo. Other notable high-grossing films include Titanic ($1.8 billion), Jurassic World ($1.6 billion) and Harry Potter And The Deathly Hallows Part 2 ($1.4 billion).


**Human**: how is cheese made? <br />
**Assistant**: Cheese making involves the process of curdling milk, separating it into a liquid and solid phase, then heating that mixture to create a semi-solid mass. The resulting product can be either hard or soft depending on its composition. Hard cheeses are typically produced from cow's milk while softer varieties come from sheep's milk.


**Human**: print all letters of the alphabet in reverse <br />
**Assistant**: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z


**Human**: write a javascript function that sorts an array using quicksort algorithm <br />
**Assistant**: 
```
function quicksort(array) {
    if (array.length < 2) {
        return array;
    }
    var pivot = array[0];
    var left = quicksort(array, 0, array.length - 1);
    var right = quicksort(array, array.length - 1, 0);
    if (left < pivot) {
        left = pivot;
    }
    if (right > pivot) {
        right = pivot;
    }
    if (left < right) {
        quicksort(array, left, right);
    }
```
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_lxe__Cerebras-GPT-2.7B-Alpaca-SP)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 29.4   |
| ARC (25-shot)         | 30.8          |
| HellaSwag (10-shot)   | 48.88    |
| MMLU (5-shot)         | 25.12         |
| TruthfulQA (0-shot)   | 40.24   |
| Winogrande (5-shot)   | 55.41   |
| GSM8K (5-shot)        | 0.53        |
| DROP (3-shot)         | 4.78         |