Update README.md
Browse files
README.md
CHANGED
@@ -5,110 +5,120 @@ language:
|
|
5 |
- en
|
6 |
pipeline_tag: text2text-generation
|
7 |
tags:
|
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 |
-
|
|
|
|
|
50 |
|
51 |
-
|
52 |
|
53 |
-
|
54 |
-
|
|
|
55 |
|
56 |
-
|
|
|
|
|
57 |
|
58 |
-
|
59 |
|
60 |
-
|
61 |
|
62 |
-
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
-
|
66 |
|
67 |
-
|
68 |
|
69 |
-
|
|
|
70 |
|
71 |
-
|
72 |
-
- **Hours used:** More information needed
|
73 |
-
- **Cloud Provider:** GCP
|
74 |
-
- **Compute Region:** More information needed
|
75 |
-
- **Carbon Emitted:** More information needed
|
76 |
|
77 |
-
|
78 |
|
79 |
-
|
80 |
|
81 |
```bibtex
|
82 |
@article{2024t5v2,
|
83 |
author = {Lintang Sutawika and Aran Komatsuzaki and Colin Raffel},
|
84 |
-
title = {
|
85 |
year = {2024},
|
86 |
url = {}
|
87 |
}
|
88 |
-
```
|
89 |
-
|
90 |
-
# How to Get Started with the Model
|
91 |
-
|
92 |
-
Use the code below to get started with the model.
|
93 |
-
|
94 |
-
<details>
|
95 |
-
<summary> Click to expand </summary>
|
96 |
-
|
97 |
-
```python
|
98 |
-
from transformers import UMT5Tokenizer, UMT5Model
|
99 |
-
|
100 |
-
tokenizer = UMT5Tokenizer.from_pretrained("EleutherAI/t5-v2-base")
|
101 |
-
model = UMT5Model.from_pretrained("EleutherAI/t5-v2-base")
|
102 |
-
|
103 |
-
input_ids = tokenizer(
|
104 |
-
"Studies have been shown that owning a dog is good for you", return_tensors="pt"
|
105 |
-
).input_ids # Batch size 1
|
106 |
-
decoder_input_ids = tokenizer("Studies show that", return_tensors="pt").input_ids # Batch size 1
|
107 |
-
|
108 |
-
# forward pass
|
109 |
-
outputs = model(input_ids=input_ids, decoder_input_ids=decoder_input_ids)
|
110 |
-
last_hidden_states = outputs.last_hidden_state
|
111 |
-
```
|
112 |
-
|
113 |
-
|
114 |
-
</details>
|
|
|
5 |
- en
|
6 |
pipeline_tag: text2text-generation
|
7 |
tags:
|
8 |
+
- t5x
|
9 |
+
- encode-decoder
|
10 |
---
|
11 |
|
12 |
+
Pile-T5 Base is an Encoder-Decoder model trained on [the Pile](https://pile.eleuther.ai/) using the [T5x](https://github.com/google-research/t5x) library. The model was trained for 2 million steps or roughly 2 trillion tokens using MLM-objective similar to the original T5 model.
|
13 |
|
14 |
+
### Model Details
|
15 |
|
16 |
+
- Developed by: [EleutherAI](http://eleuther.ai)
|
17 |
+
- Model type: Transformer-based Language Model
|
18 |
+
- Language: English
|
19 |
+
- Learn more: [Blogpost](). For details about the training dataset,
|
20 |
+
see [the Pile paper](https://arxiv.org/abs/2101.00027), and [its data
|
21 |
+
sheet](https://arxiv.org/abs/2201.07311).
|
22 |
+
- License: Apache 2.0
|
23 |
+
- Contact: to ask questions about this model, join the [EleutherAI
|
24 |
+
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`.
|
25 |
+
Please read the existing GPT-NeoX-20B documentation before asking about the model
|
26 |
+
on Discord. For general correspondence: [contact@eleuther.
|
27 |
+
ai](mailto:contact@eleuther.ai).
|
28 |
|
29 |
+
### Uses and limitations
|
30 |
|
31 |
+
#### Intended use
|
32 |
|
33 |
+
Pile-T5 was developed primarily for research purposes. It learns an inner
|
34 |
+
representation of the English language that can be used to extract features
|
35 |
+
useful for downstream tasks.
|
36 |
|
37 |
+
In addition to scientific uses, you may also further fine-tune and adapt
|
38 |
+
Pile-T5 for deployment, as long as your use is in accordance with the
|
39 |
+
Apache 2.0 license. This model works with the [Transformers
|
40 |
+
Library](https://huggingface.co/docs/transformers/index). If you decide to use
|
41 |
+
pre-trained Pile-T5 as a basis for your fine-tuned model, please note that
|
42 |
+
you need to conduct your own risk and bias assessment.
|
43 |
|
44 |
+
#### Out-of-scope use
|
45 |
|
46 |
+
Pile-T5 is **not** intended for deployment as-is. It is not a product
|
47 |
+
and cannot be used for human-facing interactions without supervision.
|
48 |
|
49 |
+
Pile-T5 has not been fine-tuned for downstream tasks for which language
|
50 |
+
models are commonly deployed, such as writing genre prose, or commercial
|
51 |
+
chatbots. This means Pile-T5 will likely **not** respond to a given prompt
|
52 |
+
the way products such as ChatGPT do. This is because, unlike Pile-T5,
|
53 |
+
ChatGPT was fine-tuned using methods such as Reinforcement Learning from Human
|
54 |
+
Feedback (RLHF) to better “understand” human instructions and dialogue.
|
55 |
|
56 |
+
This model is English-language only, and thus cannot be used for translation
|
57 |
+
or generating text in other languages.
|
58 |
|
59 |
+
#### Limitations and biases
|
60 |
|
61 |
+
The core functionality of Pile-T5 is to take a string of text that has been
|
62 |
+
partially replaced with mask tokens and predict a sequence of tokens that would
|
63 |
+
replace those mask tokens. Remember that the statistically most likely sequence
|
64 |
+
of tokens need not result in the most “accurate” text. Never rely on Pile-T5 to produce
|
65 |
+
factually accurate output.
|
66 |
|
67 |
+
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset
|
68 |
+
known to contain profanity and texts that are lewd or otherwise offensive.
|
69 |
+
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a
|
70 |
+
discussion of documented biases with regards to gender, religion, and race.
|
71 |
+
Pile-T5 may produce socially unacceptable or undesirable text, *even if*
|
72 |
+
the prompt itself does not include anything explicitly offensive.
|
73 |
|
74 |
+
We recommend curating the outputs of this model before presenting it to a human
|
75 |
+
reader. Please inform your audience that you are using artificially generated
|
76 |
+
text.
|
77 |
|
78 |
+
#### How to use
|
79 |
|
80 |
+
Pile-T5 can be loaded using the `AutoModelForSeq2SeqLM` functionality:
|
81 |
+
```python
|
82 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
83 |
|
84 |
+
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/pile-t5-base")
|
85 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("EleutherAI/pile-t5-base")
|
86 |
+
```
|
87 |
|
88 |
+
### Training
|
89 |
|
90 |
+
#### Training dataset
|
91 |
|
92 |
+
The Pile is a 825GiB general-purpose dataset in English. It was created by
|
93 |
+
EleutherAI specifically for training large language models. It contains texts
|
94 |
+
from 22 diverse sources, roughly broken down into five categories: academic
|
95 |
+
writing (e.g. arXiv), internet (e.g. CommonCrawl), prose (e.g. Project
|
96 |
+
Gutenberg), dialogue (e.g. YouTube subtitles), and miscellaneous (e.g. GitHub,
|
97 |
+
Enron Emails). See [the Pile paper](https://arxiv.org/abs/2101.00027) for
|
98 |
+
a breakdown of all data sources, methodology, and a discussion of ethical
|
99 |
+
implications. Consult [the datasheet](https://arxiv.org/abs/2201.07311) for
|
100 |
+
more detailed documentation about the Pile and its component datasets. The
|
101 |
+
Pile can be downloaded from the [official website](https://pile.eleuther.ai/),
|
102 |
+
or from a [community mirror](https://the-eye.eu/public/AI/pile/).
|
103 |
|
104 |
+
The Pile was deduplicated before being used to train Pile-T5.
|
105 |
|
106 |
+
#### Training procedure
|
107 |
|
108 |
+
Pile-T5 was trained with a batch size of approximately 1M tokens
|
109 |
+
(2048 sequences of 512 tokens each), for a total of 2,000,000 steps.
|
110 |
|
111 |
+
### Evaluations
|
|
|
|
|
|
|
|
|
112 |
|
113 |
+
TBD
|
114 |
|
115 |
+
### BibTeX
|
116 |
|
117 |
```bibtex
|
118 |
@article{2024t5v2,
|
119 |
author = {Lintang Sutawika and Aran Komatsuzaki and Colin Raffel},
|
120 |
+
title = {Pile T5, an update of T5},
|
121 |
year = {2024},
|
122 |
url = {}
|
123 |
}
|
124 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|