Skip to content

Commit 10bf13d

Browse files
CalebCourierCalebCourier
committed
Merge branch 'main' into exception-handling
2 parents 98fbfcd + 37672e5 commit 10bf13d

File tree

8 files changed

+571
-94
lines changed

8 files changed

+571
-94
lines changed

‎.github/workflows/ci.yml‎

Lines changed: 41 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -142,48 +142,47 @@ jobs:
142142

143143
Notebooks:
144144
runs-on: ubuntu-latest
145-
if: ${{ always() && contains(join(needs.*.result, ','), 'success') }}
146145
needs: [Linting, Typing, Pytests]
147-
146+
strategy:
147+
matrix:
148+
# this line is automatically generated by the script in .github/workflows/scripts/update_notebook_matrix.sh
149+
notebook: ["bug_free_python_code.ipynb","check_for_pii.ipynb","competitors_check.ipynb","extracting_entities.ipynb","generate_structured_data.ipynb","generate_structured_data_cohere.ipynb","guardrails_with_chat_models.ipynb","input_validation.ipynb","llamaindex-output-parsing.ipynb","no_secrets_in_generated_text.ipynb","provenance.ipynb","recipe_generation.ipynb","regex_validation.ipynb","response_is_on_topic.ipynb","secrets_detection.ipynb","select_choice_based_on_action.ipynb","streaming.ipynb","syntax_error_free_sql.ipynb","text_summarization_quality.ipynb","toxic_language.ipynb","translation_to_specific_language.ipynb","translation_with_quality_check.ipynb","valid_chess_moves.ipynb","value_within_distribution.ipynb"]
148150
env:
149-
COHERE_API_KEY: ${{ secrets.COHERE_API_KEY }}
150-
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
151-
HUGGINGFACE_API_KEY: ${{ secrets.HUGGINGFACE_API_KEY }}
152-
151+
COHERE_API_KEY: ${{ secrets.COHERE_API_KEY }}
152+
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
153+
HUGGINGFACE_API_KEY: ${{ secrets.HUGGINGFACE_API_KEY }}
154+
NLTK_DATA: /tmp/nltk_data
153155
steps:
154-
- name: Checkout repository
155-
uses: actions/checkout@v2
156-
157-
- name: Set up Python
158-
uses: actions/setup-python@v2
159-
with:
160-
python-version: 3.11.x
161-
162-
- name: Poetry cache
163-
uses: actions/cache@v3
164-
with:
165-
path: ~/.cache/pypoetry
166-
key: poetry-cache-${{ runner.os }}-${{ steps.setup_python.outputs.python-version }}-${{ env.POETRY_VERSION }}
167-
168-
- name: Install Poetry
169-
uses: snok/install-poetry@v1
170-
with:
171-
virtualenvs-create: true
172-
virtualenvs-in-project: true
173-
installer-parallel: true
174-
175-
- name: Install dependencies
176-
run: |
177-
make full;
178-
poetry add openai==0.28.1 jupyter nbconvert cohere;
179-
# pip install openai==0.28.1 jupyter nbconvert;
180-
# pip install .;
181-
182-
- name: Check for pypdfium2
183-
run: poetry run pip show pypdfium2
184-
185-
- name: Huggingface Hub Login
186-
run: poetry run huggingface-cli login --token $HUGGINGFACE_API_KEY
187-
188-
- name: Execute notebooks and check for errors
189-
run: ./.github/workflows/scripts/run_notebooks.sh
156+
- name: Checkout repository
157+
uses: actions/checkout@v2
158+
- name: Set up Python
159+
uses: actions/setup-python@v2
160+
with:
161+
python-version: 3.11.x
162+
- name: Poetry cache
163+
uses: actions/cache@v3
164+
with:
165+
path: ~/.cache/pypoetry
166+
key: poetry-cache-${{ runner.os }}-${{ steps.setup_python.outputs.python-version }}-${{ env.POETRY_VERSION }}
167+
- name: Install Poetry
168+
uses: snok/install-poetry@v1
169+
with:
170+
virtualenvs-create: true
171+
virtualenvs-in-project: true
172+
installer-parallel: true
173+
- name: Install dependencies
174+
run: |
175+
make full;
176+
poetry add openai==0.28.1 jupyter nbconvert cohere;
177+
- name: Check for pypdfium2
178+
run: poetry run pip show pypdfium2
179+
- name: Huggingface Hub Login
180+
run: poetry run huggingface-cli login --token $HUGGINGFACE_API_KEY
181+
- name: download nltk data
182+
run: |
183+
mkdir /tmp/nltk_data;
184+
poetry run python -m nltk.downloader -d /tmp/nltk_data punkt;
185+
- name: Use venv
186+
run: source .venv/bin/activate
187+
- name: Execute notebooks and check for errors
188+
run: bash ./.github/workflows/scripts/run_notebooks.sh ${{ matrix.notebook }}

‎.github/workflows/examples_check.yml‎

Lines changed: 18 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -8,48 +8,47 @@ on:
88
jobs:
99
execute_notebooks:
1010
runs-on: ubuntu-latest
11-
11+
strategy:
12+
matrix:
13+
# this line is automatically generated by the script in .github/workflows/scripts/update_notebook_matrix.sh
14+
notebook: ["bug_free_python_code.ipynb","check_for_pii.ipynb","competitors_check.ipynb","extracting_entities.ipynb","generate_structured_data.ipynb","generate_structured_data_cohere.ipynb","guardrails_with_chat_models.ipynb","input_validation.ipynb","llamaindex-output-parsing.ipynb","no_secrets_in_generated_text.ipynb","provenance.ipynb","recipe_generation.ipynb","regex_validation.ipynb","response_is_on_topic.ipynb","secrets_detection.ipynb","select_choice_based_on_action.ipynb","streaming.ipynb","syntax_error_free_sql.ipynb","text_summarization_quality.ipynb","toxic_language.ipynb","translation_to_specific_language.ipynb","translation_with_quality_check.ipynb","valid_chess_moves.ipynb","value_within_distribution.ipynb"]
1215
env:
1316
COHERE_API_KEY: ${{ secrets.COHERE_API_KEY }}
1417
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
1518
HUGGINGFACE_API_KEY: ${{ secrets.HUGGINGFACE_API_KEY }}
16-
17-
19+
NLTK_DATA: /tmp/nltk_data
1820
steps:
1921
- name: Checkout repository
20-
uses: actions/checkout@v2
21-
22+
uses: actions/checkout@v2
2223
- name: Set up Python
2324
uses: actions/setup-python@v2
24-
with:
25+
with:
2526
python-version: 3.11.x
26-
2727
- name: Poetry cache
2828
uses: actions/cache@v3
2929
with:
3030
path: ~/.cache/pypoetry
3131
key: poetry-cache-${{ runner.os }}-${{ steps.setup_python.outputs.python-version }}-${{ env.POETRY_VERSION }}
32-
3332
- name: Install Poetry
3433
uses: snok/install-poetry@v1
3534
with:
36-
virtualenvs-create: true
37-
virtualenvs-in-project: true
38-
installer-parallel: true
39-
35+
virtualenvs-create: true
36+
virtualenvs-in-project: true
37+
installer-parallel: true
4038
- name: Install dependencies
4139
run: |
4240
make full;
4341
poetry add openai==0.28.1 jupyter nbconvert cohere;
44-
# pip install openai==0.28.1 jupyter nbconvert;
45-
# pip install .;
46-
4742
- name: Check for pypdfium2
4843
run: poetry run pip show pypdfium2
49-
5044
- name: Huggingface Hub Login
51-
run: huggingface-cli login --token $HUGGINGFACE_API_KEY
52-
45+
run: poetry run huggingface-cli login --token $HUGGINGFACE_API_KEY
46+
- name: download nltk data
47+
run: |
48+
mkdir /tmp/nltk_data;
49+
poetry run python -m nltk.downloader -d /tmp/nltk_data punkt;
50+
- name: Use venv
51+
run: source .venv/bin/activate
5352
- name: Execute notebooks and check for errors
54-
run: ./.github/workflows/scripts/run_notebooks.sh
53+
run: bash ./.github/workflows/scripts/run_notebooks.sh ${{ matrix.notebook }}
5554

‎.github/workflows/scripts/run_notebooks.sh‎

Lines changed: 13 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -1,38 +1,20 @@
11
#!/bin/bash
2-
3-
mkdir /tmp/nltk_data;
4-
poetry run python -m nltk.downloader -d /tmp/nltk_data punkt;
52
export NLTK_DATA=/tmp/nltk_data;
63

74
cd docs/examples
85

9-
# Function to process a notebook
10-
process_notebook() {
11-
notebook="$1"
12-
invalid_notebooks=("valid_chess_moves.ipynb" "llamaindex-output-parsing.ipynb" "competitors_check.ipynb")
13-
if [[ ! " ${invalid_notebooks[@]} " =~ " ${notebook} " ]]; then
14-
echo "Processing $notebook..."
15-
poetry run jupyter nbconvert --to notebook --execute "$notebook"
16-
if [ $? -ne 0 ]; then
17-
echo "Error found in $notebook"
18-
echo "Error in $notebook. See logs for details." >> errors.txt
19-
fi
20-
fi
21-
}
22-
23-
export -f process_notebook # Export the function for parallel execution
24-
25-
# Create a file to collect errors
26-
> errors.txt
6+
# Get the notebook name from the matrix variable
7+
notebook="$1"
278

28-
# Run in parallel
29-
ls *.ipynb | parallel process_notebook
9+
# Check if the notebook should be processed
10+
invalid_notebooks=("valid_chess_moves.ipynb" "llamaindex-output-parsing.ipynb" "competitors_check.ipynb")
11+
if [[ ! " ${invalid_notebooks[@]} " =~ " ${notebook} " ]]; then
12+
echo "Processing $notebook..."
13+
poetry run jupyter nbconvert --to notebook --execute "$notebook"
14+
if [ $? -ne 0 ]; then
15+
echo "Error found in $notebook"
16+
echo "Error in $notebook. See logs for details." >> errors.txt
17+
fi
18+
fi
3019

31-
# Check if there were any errors
32-
if [ -s errors.txt ]; then
33-
echo "Some notebooks had errors"
34-
cat errors.txt
35-
exit 1
36-
else
37-
echo "All notebooks ran successfully."
38-
fi
20+
exit 0

‎.github/workflows/scripts/update_notebook_matrix.sh‎

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
#!/bin/bash
2+
# Array to store notebook names
3+
notebook_names="["
4+
5+
# Compile list of file names
6+
for file in $(ls docs/examples/*.ipynb); do
7+
# Add the full filename with extension
8+
filename=$(basename "$file")
9+
10+
notebook_names+="\"$filename\","
11+
done
12+
notebook_names="${notebook_names%,}]"
13+
14+
# echo $notebook_names
15+
16+
17+
# find line that begins with "notebook:" and replace it with notebook: $notebook_names
18+
sed "s/notebook: \[.*\]/notebook: $notebook_names/" .github/workflows/examples_check.yml > .github/workflows/examples_check.yml.tmp
19+
mv .github/workflows/examples_check.yml.tmp .github/workflows/examples_check.yml
20+
21+
sed "s/notebook: \[.*\]/notebook: $notebook_names/" .github/workflows/ci.yml > .github/workflows/ci.yml.tmp
22+
mv .github/workflows/ci.yml.tmp .github/workflows/ci.yml

‎Makefile‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -71,3 +71,4 @@ precommit:
7171
# pytest -x -q --no-summary
7272
pyright guardrails/
7373
make lint
74+
./github/workflows/scripts/update_notebook_matrix.sh

‎docs/llm_api_wrappers.md‎

Lines changed: 103 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -86,7 +86,7 @@ guard = gd.Guard.from_rail(...)
8686
anthropic_client = Anthropic(api_key="my_api_key")
8787

8888
# Wrap Anthropic API call
89-
raw_llm_output, guardrail_output = guard(
89+
raw_llm_output, guardrail_output, *rest = guard(
9090
anthropic_client.completions.create,
9191
prompt_params={
9292
"prompt_param_1": "value_1",
@@ -100,6 +100,108 @@ raw_llm_output, guardrail_output = guard(
100100
```
101101

102102

103+
## Hugging Face
104+
105+
### Text Generation Models
106+
```py
107+
from guardrails import Guard
108+
from guardrails.validators import ValidLength, ToxicLanguage
109+
import torch
110+
from transformers import AutoModelForCausalLM, AutoTokenizer
111+
112+
113+
# Create your prompt or starting text
114+
prompt = "Hello, I'm a language model,"
115+
116+
# Setup torch
117+
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
118+
119+
# Instantiate your tokenizer
120+
tokenizer = AutoTokenizer.from_pretrained("gpt2")
121+
122+
# Instantiate your model
123+
model = AutoModelForCausalLM.from_pretrained("gpt2", pad_token_id=tokenizer.eos_token_id).to(torch_device)
124+
125+
# Customize your model inputs if desired.
126+
# If you don't pass and inputs (`input_ids`, `input_values`, `input_features`, or `pixel_values`)
127+
# We'll try to do something similar to below using the tokenizer and the prompt.
128+
# We strongly suggest passing in your own inputs.
129+
model_inputs = tokenizer(prompt, return_tensors="pt").to(torch_device)
130+
131+
132+
# Create the Guard
133+
guard = Guard.from_string(
134+
validators=[
135+
ValidLength(
136+
min=48,
137+
on_fail="fix"
138+
),
139+
ToxicLanguage(
140+
on_fail="fix"
141+
)
142+
],
143+
prompt=prompt
144+
)
145+
146+
# Run the Guard
147+
response = guard(
148+
llm_api=model.generate,
149+
max_new_tokens=40,
150+
tokenizer=tokenizer,
151+
**model_inputs,
152+
)
153+
154+
# Check the output
155+
if response.validation_passed:
156+
print("validated_output: ", response.validated_output)
157+
else:
158+
print("error: ", response.error)
159+
160+
```
161+
162+
### Pipelines
163+
```py
164+
from guardrails import Guard
165+
from guardrails.validators import ValidLength, ToxicLanguage
166+
import torch
167+
from transformers import pipeline
168+
169+
170+
# Create your prompt or starting text
171+
prompt = "What are we having for dinner?"
172+
173+
# Setup pipeline
174+
generator = pipeline("text-generation", model="facebook/opt-350m")
175+
176+
177+
# Create the Guard
178+
guard = Guard.from_string(
179+
validators=[
180+
ValidLength(
181+
min=48,
182+
on_fail="fix"
183+
),
184+
ToxicLanguage(
185+
on_fail="fix"
186+
)
187+
],
188+
prompt=prompt
189+
)
190+
191+
# Run the Guard
192+
response = guard(
193+
llm_api=generator,
194+
max_new_tokens=40
195+
)
196+
197+
if response.validation_passed:
198+
print("validated_output: ", response.validated_output)
199+
else:
200+
print("error: ", response.error)
201+
202+
```
203+
204+
103205
## Using Manifest
104206
[Manifest](https://github.com/HazyResearch/manifest) is a wrapper around most model APIs and supports hosting local models. It can be used as a LLM API.
105207

0 commit comments

Comments
 (0)