Observed since 21 Oct, 09:04 UTC
No superseded metadata states collected. Unchanged observations do not create new revisions.
agi
A Python client library for interacting with the General Reasoning platform
A Python client library for interacting with the General Reasoning platform API.
The documentation is available here.
You can install the package using pip:
pip install agi
Obtain an API key from the website. Then:
import agi
client = agi.Client(api_key=YOUR_API_KEY)
# Download reasoning traces and verifications as a .jsonl
client.data.get(task='math-word-problems', model='DeepSeek-R1')
The full evaluation guide is available here. A boilerplate example is shown below:
import agi
client = agi.Client("YOUR_API_KEY")
model = "USERNAME/MODEL_NAME"
# Retrieve test questions
data = client.evals.get(
task='mathematical-brainteasers',
split='test'
)
question = data['questions']
# Submit model evaluations
for question in questions:
reasoning_trace, answer = MyModel(
system_prompt=question['system_prompt'],
question=question['text']
) # Replace with your model logic
client.evals.submit(
id=question['id'],
model=model,
reasoning_trace=reasoning_trace,
answer=answer
)
Read more about our grader and verifier here
import agi
client = agi.Client(api_key=YOUR_API_KEY)
response = client.verify.create(
model="endeavour-preview",
question="Are Dyson spheres feasible to build?",
model_answer="Yes"
)
import agi
client = agi.Client(api_key=YOUR_API_KEY)
response = client.grade.create(
model="endeavour-preview",
question="Are Dyson spheres feasible to build?",
model_answer="Yes",
ground_truth_answer="No",
)
This project is licensed under the MIT License - see the LICENSE file for details.
Download counts and popularity will appear here as the collector receives them.
| Package / target | Requirement | Relationship |
|---|---|---|
requestsEnvironment{
"extras": "",
"marker": null,
"requires_python": ">=3.7",
"evaluation": "NOT_EVALUATED"
} | requests>=2.31.0 | runtime |
bdist_wheel · 4.81 KB · Python py3
sdist · 4.78 KB · Python source
Observed since 21 Oct, 09:04 UTC
No superseded metadata states collected. Unchanged observations do not create new revisions.
package_version:070834e15f014effddb63d8142f167f7eb4b90e5210bb1f234873310a2ea07f7