---
language:
  - en
tags:
  - pytorch
  - causal-lm
  - pythia
license: apache-2.0
datasets:
  - Dahoas/synthetic-instruct-gptj-pairwise
---

This model is created by finetuning [`EleutherAI/pythia-6.9b-deduped`](https://huggingface.co/EleutherAI/pythia-6.9b-deduped) on the [`Dahoas/synthetic-instruct-gptj-pairwise`](https://huggingface.co/datasets/Dahoas/synthetic-instruct-gptj-pairwise).

You can try a [demo](https://cloud.lambdalabs.com/demos/ml/sft-pythia-side-by-side) of the model hosted on [Lambda Cloud](https://lambdalabs.com/service/gpu-cloud).

### Model Details

- Finetuned by: [Lambda](https://lambdalabs.com/)
- Model type: Transformer-based Language Model
- Language: English
- Pre-trained model: [EleutherAI/pythia-6.9b-deduped](https://huggingface.co/EleutherAI/pythia-6.9b-deduped)
- Dataset: [Dahoas/synthetic-instruct-gptj-pairwise](https://huggingface.co/datasets/Dahoas/synthetic-instruct-gptj-pairwise)
- Library: [transformers](https://huggingface.co/docs/transformers/index)
- License: Apache 2.0

### Prerequisites

Running inference with the model takes ~17GB of GPU memory.

### Quick Start

```
import torch

from transformers import AutoTokenizer, pipeline, StoppingCriteria, StoppingCriteriaList

device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")

model_name = "lambdalabs/pythia-6.9b-deduped-synthetic-instruct"
max_new_tokens = 1536
stop_token = "<|stop|>"


class KeywordsStoppingCriteria(StoppingCriteria):
    def __init__(self, keywords_ids: list):
        self.keywords = keywords_ids

    def __call__(
        self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs
    ) -> bool:
        if input_ids[0][-1] in self.keywords:
            return True
        return False


tokenizer = AutoTokenizer.from_pretrained(
    model_name,
)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.add_tokens([stop_token])

stop_ids = [tokenizer.encode(w)[0] for w in [stop_token]]
stop_criteria = KeywordsStoppingCriteria(stop_ids)

generator = pipeline(
    "text-generation",
    model=model_name,
    device=device,
    max_new_tokens=max_new_tokens,
    torch_dtype=torch.float16,
    stopping_criteria=StoppingCriteriaList([stop_criteria]),
)

example = "How can I make an omelette."
text = "Question: {}\nAnswer:".format(example)

result = generator(
    text,
    num_return_sequences=1,
)

output = result[0]["generated_text"]

print(output)
```

Output:

```

Question: How can I make an omelette.
Answer:To make an omelette, start by gathering the ingredients you will need. Beat some eggs in a bowl and season with salt and pepper. Heat a non-stick pan over medium heat and add a tablespoon of butter. Once the butter has melted, pour in the egg mixture and let it cook for a few minutes. As it cooks, use a spatula to lift the edges of the omelette and tilt the pan so that the uncooked egg runs underneath. Once the eggs are mostly cooked, add your desired fillings and fold the omelette in half. Let it cook for a few more minutes, then slide it onto a plate and enjoy.<|stop|>

```

### Training

The model was trained on the [`Dahoas/synthetic-instruct-gptj-pairwise`](https://huggingface.co/datasets/Dahoas/synthetic-instruct-gptj-pairwise). We split the original dataset into the train (first 32000 examples) and validation (the remaining 1144 examples) subsets.

We finetune the model for 4 epoches with the help of deepspeed. This took 8xA100 80GB 6 hours, where we set `batch_size_per_gpu` to `8` (so global batch size is 64), and learning rate to `0.000005` (with linear decay to zero at the last trainig step). You can find a Weights and Biases record [here](https://wandb.ai/chuanli11/ft-synthetic-instruct-gptj-pairwise-pythia6.9b-deepspeed?workspace=user-).