|
--- |
|
license: mit |
|
datasets: |
|
- databricks/databricks-dolly-15k |
|
language: |
|
- en |
|
pipeline_tag: text-generation |
|
tags: |
|
- dolly |
|
- dolly-v2 |
|
- instruct |
|
- sharded |
|
- quantized |
|
inference: False |
|
--- |
|
|
|
|
|
# dolly-v2-7b: **8-bit** sharded checkpoint |
|
|
|
|
|
<a href="https://colab.research.google.com/gist/pszemraj/8100e98caab538be32832d1208e93f65/dolly-v2-7b-8bit-inference.ipynb"> |
|
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/> |
|
</a> |
|
|
|
This is a sharded checkpoint (with ~2GB shards) of the `databricks/dolly-v2-7b` model **in 8-bit precision using `bitsandbytes`**. |
|
|
|
Refer to the [original model](https://huggingface.co/databricks/dolly-v2-7b) for all details. For more info on loading 8bit models, refer to the [example repo](https://huggingface.co/ybelkada/bloom-1b7-8bit) and/or the `4.28.0` [release info](https://github.com/huggingface/transformers/releases/tag/v4.28.0). |
|
|
|
- total model size is only ~7.5 GB! |
|
- this enables low-RAM loading, i.e. Colab :) |
|
|
|
## Basic Usage |
|
|
|
|
|
install/upgrade `transformers`, `accelerate`, and `bitsandbytes`. For this to work **you must have** `transformers>=4.28.0` and `bitsandbytes>0.37.2`. |
|
|
|
```bash |
|
pip install -U -q transformers bitsandbytes accelerate |
|
``` |
|
|
|
Load the model. As it is serialized in 8bit you don't need to do anything special: |
|
|
|
```python |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
model_name = "ethzanalytics/dolly-v2-7b-sharded-8bit" |
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
|
model = AutoModelForCausalLM.from_pretrained(model_name) |
|
``` |
