1
! pip install pandas requests rich torch transformers

1
import pandas
2
pandas.set_option("display.max_rows", 10)

1
import getpass
2
import json
3
import requests
4
5
6
def distil_bearer_token(DL_USERNAME: str, DL_PASSWORD: str) -> str:
7
    response = requests.post(
8
        "https://cognito-idp.eu-central-1.amazonaws.com",
9
        headers={
10
            "X-Amz-Target": "AWSCognitoIdentityProviderService.InitiateAuth",
11
            "Content-Type": "application/x-amz-json-1.1",
12
        },
13
        data=json.dumps({
14
            "AuthParameters": {
15
                "USERNAME": DL_USERNAME,
16
                "PASSWORD": DL_PASSWORD,
17
            },
18
            "AuthFlow": "USER_PASSWORD_AUTH",
19
            "ClientId" : "4569nvlkn8dm0iedo54nbta6fd",
20
        })
21
    )
22
    response.raise_for_status()
23
    return response.json()["AuthenticationResult"]["AccessToken"]
24
25
26
DL_USERNAME = "YOUR_EMAIL"
27
DL_PASSWORD = getpass.getpass()
28
29
AUTH_HEADER = {"Authorization": distil_bearer_token(DL_USERNAME, DL_PASSWORD)}
30
print("Success")

1
from pprint import pprint
2
3
# Register a model
4
data = {"name": "testmodel-1234"}
5
response = requests.post(
6
    "https://api.distillabs.ai/models",
7
    data=json.dumps(data),
8
    headers={"content-type": "application/json", **AUTH_HEADER},
9
)
10
pprint(response.json())
11
model_id = response.json()["id"]
12
print(f"Registered a model with ID={model_id}")

1
from pprint import pprint
2
3
# Retrieve all models
4
response = requests.get(
5
    "https://api.distillabs.ai/models",
6
    headers=AUTH_HEADER
7
)
8
pprint(response.json())

1
from pathlib import Path
2
data_location = Path("data-banking-routing")
3
assert data_location.exists()

1
import json
2
import rich.json
3
4
with open(data_location.joinpath("job_description.json")) as fin:
5
    rich.print(rich.json.JSON(fin.read()))

1
from pathlib import Path
2
from IPython.display import display
3
4
import pandas
5
6
print("# --- Train set")
7
train = pandas.read_csv(data_location.joinpath("train.csv"))
8
display(train)
9
10
print("# --- Test set")
11
test = pandas.read_csv(data_location.joinpath("test.csv"))
12
display(test)

1
unstructured = pandas.read_csv(data_location.joinpath("unstructured.csv"))
2
display(unstructured)

1
import json
2
3
import requests
4
import yaml
5
from pathlib import Path
6
import pandas
7
8
# Specify the config
9
config = {
10
    "base": {
11
        "task": "classification",
12
    }
13
}
14
15
# Package your data
16
data_dir = Path("data")
17
data = {
18
    "job_description": {
19
        "type": "json",
20
        "content": open(data_location / "job_description.json", encoding="utf-8").read()
21
    },
22
    "train_data": {
23
        "type": "csv",
24
        "content": open(data_location / "train.csv", encoding="utf-8").read()
25
    },
26
    "test_data": {
27
        "type": "csv",
28
        "content": open(data_location / "test.csv", encoding="utf-8").read()
29
    },
30
    "unstructured_data": {
31
        "type": "csv",
32
        "content": open(data_location / "unstructured.csv", encoding="utf-8").read()
33
    },
34
    "config": {
35
        "type": "yaml",
36
        "content": yaml.dump(config)
37
    },
38
}
39
40
# Upload data
41
response = requests.post(
42
    f"https://api.distillabs.ai/models/{model_id}/uploads",
43
    data=json.dumps(data),
44
    headers={"content-type": "application/json", **AUTH_HEADER},
45
)
46
print(response.json())
47
upload_id = response.json()["id"]

1
from pprint import pprint
2
3
# Start teacher evaluation
4
data = {"upload_id": upload_id}
5
response = requests.post(
6
    f"https://api.distillabs.ai/models/{model_id}/teacher-evaluations",
7
    data=json.dumps(data),
8
    headers={"content-type": "application/json", **AUTH_HEADER},
9
)
10
11
pprint(response.json())
12
teacher_evaluation_id = response.json().get("id")

1
import json
2
from pprint import pprint
3
import pandas as pd
4
5
response = requests.get(
6
    f"https://api.distillabs.ai/teacher-evaluations/{teacher_evaluation_id}/status",
7
    headers=AUTH_HEADER,
8
)
9
pprint(response.json()["message"])
10
11
try:
12
    display(pd.DataFrame(response.json().get("results")).transpose())
13
except:
14
    pass

1
import time
2
from pprint import pprint
3
4
# Start SLM training
5
data = {"upload_id": upload_id}
6
response = requests.post(
7
    f"https://api.distillabs.ai/models/{model_id}/training",
8
    data=json.dumps(data),
9
    headers={"content-type": "application/json", **AUTH_HEADER},
10
)
11
12
pprint(response.json())
13
slm_training_job_id = response.json().get("id")

1
import json
2
3
response = requests.get(
4
    f"https://api.distillabs.ai/trainings/{slm_training_job_id}/status",
5
    headers=AUTH_HEADER,
6
)
7
response.json()

1
from pprint import pprint
2
response = requests.get(
3
    f"https://api.distillabs.ai/trainings/{slm_training_job_id}/evaluation-results",
4
    headers=AUTH_HEADER,
5
)
6
7
pprint(response.json()["message"])
8
for model in ["teacher", "student-base", "student-tuned"]:
9
    try:
10
        print(model)
11
        print("Accuracy:", response.json().get("evaluation_results").get(model).get("accuracy"))
12
        display(pd.DataFrame(response.json().get("evaluation_results").get(model)).transpose())
13
    except:
14
        pass

1
import json
2
from pprint import pprint
3
import requests
4
5
response = requests.get(
6
    f"https://api.distillabs.ai/models",
7
    headers=AUTH_HEADER,
8
)
9
pprint(response.json())

1
from pprint import pprint
2
3
slm_training_job_id = "SELECTED-MODEL"
4
response = requests.get(
5
    f"https://api.distillabs.ai/trainings/{slm_training_job_id}/model",
6
    headers=AUTH_HEADER,
7
)
8
pprint(response.json())

1
import tarfile
2
import urllib.request
3
4
print("Downloading …")
5
def status(count, block, total):
6
    print("\r", f"Downloading: {count * block / total:.1%}", end="")
7
8
9
urllib.request.urlretrieve(
10
    s3url,
11
    "model.tar",
12
    reporthook=status,
13
)
14
15
print("\nUnpacking …")
16
with tarfile.open("model.tar", mode="r:*") as tar:
17
    tar.extractall(path=".")

1
!ls -lt

1
import torch
2
import pandas
3
4
from transformers import AutoTokenizer, AutoModelForSequenceClassification
5
from transformers import TextClassificationPipeline
6
7
model = AutoModelForSequenceClassification.from_pretrained("model")
8
tokenizer = AutoTokenizer.from_pretrained("model", padding_side="left")
9
llm = TextClassificationPipeline(model=model, tokenizer=tokenizer, top_k=None)

1
answer = llm("I have a charge for cash withdrawal that I want to learn about")
2
pandas.DataFrame(answer.pop()).sort_values(by="score", ascending=False)