Zero-shot text classification using autoregressive language models.
Project description
CAPPr: zero-shot text classification using autoregressive language models
Perform zero-shot text classification by estimating the probability that an inputted completion comes after an inputted prompt. Hence the name:
Completion
After
Prompt
Probability
The method is fleshed out in my question on CrossValidated.
Usage
Use a model from the OpenAI API
Specifically, this model must be compatible with the /v1/completions endpoint.
Let's classify this sentiment example from the OpenAI text completion docs.
from cappr.openai.classify import predict
tweet = 'I loved the new Batman movie!'
prompt = f'Tweet: {tweet}\nSentiment:'
class_names = ('positive', 'neutral', 'negative')
# optional: let's supply a prior distribution over the classes
prior = ( 1/8 , 1/8 , 3/4 )
preds = predict(prompts=[prompt],
completions=class_names,
model='text-ada-001',
prior=prior)
preds
# ['positive']
Use a model from the HuggingFace model hub
Specifically, this model must be able to be loaded using
transformers.AutoModelForCausalLM.from_pretrained(model)
.
Smaller LMs may not work well. But there will likely be better ones in the hub soon.
from cappr.huggingface.classify import predict
prompt = 'Which planet is closer to the Sun: Mercury or Earth?'
class_names = ('Mercury', 'Earth')
prior = None # uniform prior
preds = predict(prompts=[prompt],
completions=class_names,
model='gpt2',
prior=prior)
preds
# ['Mercury']
Run in batches
Let's use huggingface
for this example cuz it's free. And let's predict probabilities
instead of the class.
from cappr.huggingface.classify import predict_proba
prompts = [
'Stephen Curry is a',
'Martina Navratilova was a',
"Dexter, from the TV Series Dexter's Laboratory, is a",
'LeBron James is a',
]
# each of the prompts could be completed with one of these:
class_names = (
'basketball player',
'tennis player',
'scientist'
)
prior = (
1/6, # few
1/6, # few
2/3 # there are more
)
pred_probs = predict_proba(prompts=prompts,
completions=class_names,
model='gpt2',
batch_size=32, # whatever fits on your CPU/GPU
prior=prior)
# pred_probs[i,j] = probability that prompts[i] is classified as class_names[j]
print(pred_probs.round(1))
# [[0.5 0.3 0.2]
# [0.3 0.6 0.2]
# [0.1 0.1 0.8]
# [0.8 0.2 0. ]]
# for each prompt, which completion is most likely?
pred_class_idxs = pred_probs.argmax(axis=1)
print([class_names[pred_class_idx] for pred_class_idx in pred_class_idxs])
# ['basketball player',
# 'tennis player',
# 'scientist',
# 'basketball player']
Run in batches, where each prompt has a different set of possible completions
Again, let's use huggingface
to predict probabilities. And this time, let's pass in an
instantiated model and tokenizer instead of its name. That way, the model isn't
re-loaded every time you wanna use it.
import numpy as np
from transformers import AutoModelForCausalLM, AutoTokenizer
from cappr import Example
from cappr.huggingface.classify import predict_proba_examples
examples = [
Example(prompt='Jodie Foster played',
completions=('Clarice Starling', 'Trinity in The Matrix')),
Example(prompt='Batman, from Batman: The Animated Series, was played by',
completions=('Pete Holmes', 'Kevin Conroy', 'Spongebob!'),
prior= ( 1/3 , 2/3 , 0 ))
]
model_name = 'gpt2'
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
pred_probs = predict_proba_examples(examples,
model_and_tokenizer=(model, tokenizer))
# pred_probs[i][j] = probability that examples[i].prompt is classified as
# examples[i].completions[j]
print([example_pred_probs.round(2)
for example_pred_probs in pred_probs])
# [array([0.7, 0.3]),
# array([0.03, 0.97, 0. ])]
# for each example, which completion is most likely?
pred_class_idxs = [np.argmax(example_pred_probs)
for example_pred_probs in pred_probs]
print([example.completions[pred_class_idx]
for example, pred_class_idx in zip(examples, pred_class_idxs)])
# ['Clarice Starling',
# 'Kevin Conroy']
More examples are linked here in the documentation.
See
demos/superglue/copa.ipynb
for a demonstration of a slightly harder classification task.
Documentation
https://cappr.readthedocs.io/en/latest/
Please let me know if you find the writing too dense. The main motivation behind this project is simplicity :-)
Setup
If you intend on using OpenAI models, sign up for the OpenAI API
here, and then set the environment variable
OPENAI_API_KEY
. For zero-shot classification, OpenAI models are currently far ahead of
others. But using them will cost ya 💰!
Install with pip
:
python -m pip install cappr
(Optional) Install requirements for HuggingFace models
python -m pip install cappr[hf]
(Optional) Install requirements for running demos
python -m pip install cappr[demos]
Motivation
Create a more usable zero-shot text classification interface than classification via sampling (CVS).
Short
In CVS, your job is to write up your classification task in a prompt
string, and then
write custom code to post-process arbitrary completion
/output strings.
In CAPPr, your job starts and stops at writing up your classification task as a
{prompt}{end_of_prompt}{completion}
string.
Long
Please see this page of the documentation.
Unstudied
I'm curious to see how much easier estimation/discrimination is than generation. In
demos/superglue/copa.ipynb
,
CVS using OpenAI's text-curie-001
is less than 50% accurate, while CAPPr is 80%
accurate.
Honest
Keep myself busy
Results
Statistical performance
Not too shabby. TODO: summary table comparing CVS vs. CAPPr vs. few-shot methods like SetFit and PET.
Computational performance
One concern was that CAPPr requires as many model()
calls as there are classes. But in
the CAPPr scheme, we can simply cache each attention block's keys and values for the
prompts. This feature is already supported by AutoModelForCausalLM
s. See this
code for
the implementation. Note that this caching is not implemented for OpenAI models, as I
can't control their backend. This means that when running cappr.openai
functions,
you'll be on the cappr (slow) line :-(
Figure 1: COPA dataset, repeating the
choices to simulate multi-class classification tasks. GPT-2
(small) was run on a Tesla K80 GPU (whatever was free in
Google Colab in March 2023, I'm not hardware savvy). 96 classification inputs were
processed in batches of size 32. Each point in the graph is a median of 5 runs. For
classification via sampling (CVS), exactly 4 tokens were generated for each prompt,
which is the number of tokens in '\n\nAnswer A'
. 1-token times are also shown. But for
COPA (and other multiple-choice style prompts), that may result in lower zero-shot
accuracy, as most of the sampled choices come after the first token.
Related work
While benchmarking this method on the Winograd Schema Challenge, I found that this paper is very similar:
Trinh, Trieu H., and Quoc V. Le. "A simple method for commonsense reasoning." arXiv preprint arXiv:1806.02847 (2018).
PET with multiple masks also aggregates token probabilities to do prompt-completion classification, but these probabilities are assumed to come from masked language models like BERT.
Schick, Timo, and Hinrich Schütze. "It's not just size that matters: Small language models are also few-shot learners." arXiv preprint arXiv:2009.07118 (2020).
Contributing
TODO
Testing
Setup
-
Clone the repo
git clone https://github.com/kddubey/cappr.git
-
Create a new Python 3.8+ environment
-
Install this package in editable mode, along with development requirements
python -m pip install -e .[dev]
Run tests
pytest
Dumping VS code extensions for development:
- autoDocstring. Use the numpy format.
- Set Python formatting to
black
. - Rewrap. Enable Auto Wrap.
- TOML Language Support
Todo
(**) = I'm currently working on this or will work on it really soon
Code
- Testing
- Increase test cases
- Some more standardization b/t openai and huggingface tests
- Add code coverage badge to look cool
- Test input checks
- Small CPU speed-ups
- For constant-completions input, vectorize
agg_log_probs
- For
examples
input, if # completions per prompt is constant, vectorizeposterior_prob
- For constant-completions input, vectorize
- Make progress bars optional, since inference often isn't batched
- Factor out input checks (on prompts and completions)
- De-automate overzealous auto-docstring stuff :-(
- HuggingFace
transformers.AutoModelForCausalLM
- Optimize backend to enable greater scaling wrt # completions/classes
- Get it working on single-GPU, check that it's faster than sampling assuming
batching
- Get to the bottom of why it's slower w/o batching
- Allow non-
' '
end_of_prompt
! I'll have to go back to the drawing board I think - Consider batchifying the completions again, since they technically don't go in
batches of
batch_size
; the actual batch size is the sum of the number of completions corresponding to the batch of prompts! Not a huge memory issue I think b/c completions are usually half as long. But it should be configurable at the very least. - Factor out repeated code b/t
classify
andclassify_no_cache
- Support Inference Endpoints?
- Support TensorFlow models if it's easy
- Support priming, as in: cache it
- (for me) Auto-enforced code formatting b/c it's getting time-consuming
- Allow for multi-label classification
- Pass
normalize
as an argument to predict_proba functions - For
huggingface
, add note that you'll get faster results by passing all labels at once (assuming prompt is identical for each label)
- Pass
- Create a notebook template
- Fill in missing or non-numpy docstrings
Research
Evaluate on more datasets, and understand its relative advantages and disadvantages vs other classification methods.
- RAFT benchmark (**)
- Zero-shot training scores
- Submit zero-shot test predictions
- Few-shot (priming) training scores
- Submit few-shot test predictions
- Create a user guide, build a table of results comparing competing approaches on statistical performance, cost, and computation
- Make a computational comparison to sampling (**)
- Assume I have full freedom to decide how inference works. Demo w/ GPT-2. Process inputs in batches.
- Process inputs 1-by-1
- More SuperGLUE tasks?
- Re-run COPA demo w/ left-stripped completions (there are a few which aren't)
- Calibration
- Is the prior actually effective? Downsample and see
- curves
- Compare against few-shot embeddings
- Finetune smaller, cheaper model and compare against zero-shot w/ davinci
- e.g., GPT-2 from huggingface,
text-ada-001
- Again, compare against sampling
- e.g., GPT-2 from huggingface,
- Evaluate a bigger model like GPT-J
- Evaluate different aggregation functions. Currently taking mean, but there was no good theory for that
- A bit ambitious: support insertion and backwards-completion. Quite ambitious b/c manipulating position IDs isn't sufficient (I think).
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.