deploy model example

Author: eff-kay

CHANGELOG.md

@@ -2,6 +2,11 @@
 All notable changes to `slashml-python-client` aka `slashml` will be documented in this file.
 This project adheres to [Semantic Versioning](https://semver.org/).
 
+## 0.1.3 - 2023-06-09
+
+### Added
+- Model deployment
+
 ## 0.1.3 - 2023-05-13
 
 ### Added

README.md

@@ -65,6 +65,50 @@ print(f"Summary = {response.summarization_data}")
 ```
 
 
+#### Deploy your own Model
+<!-- write a code snippet in the minimum number of lines  -->
+
+```python
+from slashml import ModelDeployment
+import time
+
+# you might have to install transfomers and torch
+from transformers import pipeline
+
+def train_model():
+    # Bring in model from huggingface
+    return pipeline('fill-mask', model='bert-base-uncased')
+
+my_model = train_model()
+
+# Replace `API_KEY` with your SlasML API token.
+API_KEY = "YOUR_API_KEY"
+
+model = ModelDeployment(api_key=API_KEY)
+
+# deploy model
+response = model.deploy(model_name='my_model_3', model=my_model)
+
+# wait for it to be deployed
+time.sleep(2)
+status = model.status(model_version_id=response.id)
+
+while status.status != 'READY':
+    print(f'status: {status.status}')
+    print('trying again in 5 seconds')
+    time.sleep(5)
+    status = model.status(model_version_id=response.id)
+
+    if status.status == 'FAILED':
+        raise Exception('Model deployment failed')
+
+# submit prediction
+input_text = 'Steve jobs is the [MASK] of Apple.'
+prediction = model.predict(model_version_id=response.id, model_input=input_text)
+print(prediction)
+```
+
+
 ### View the list of service providers available
 ```python
 from slashml import TextToSpeech

examples/deploy_model/deploy_hugging_face_transformer.py

@@ -0,0 +1,36 @@
+from slashml import ModelDeployment
+import time
+
+# you might have to install transfomers and torch
+from transformers import pipeline
+
+def train_model():
+    # Bring in model from huggingface
+    return pipeline('fill-mask', model='bert-base-uncased')
+
+my_model = train_model()
+
+# Replace `API_KEY` with your SlasML API token.
+API_KEY = "YOUR_API_KEY"
+
+model = ModelDeployment(api_key=API_KEY)
+
+# deploy model
+response = model.deploy(model_name='my_model_3', model=my_model)
+
+# wait for it to be deployed
+time.sleep(2)
+status = model.status(model_version_id=response.id)
+
+while status.status != 'READY':
+    print(f'status: {status.status}')
+    print('trying again in 5 seconds')
+    time.sleep(5)
+    status = model.status(model_version_id=response.id)
+
+    if status.status == 'FAILED':
+        raise Exception('Model deployment failed')
+
+# submit prediction
+input_text = 'Steve jobs is the [MASK] of Apple.'
+prediction = model.predict(model_version_id=response.id, model_input=input_text)

examples/deploy_model/requirements.txt

@@ -0,0 +1,2 @@
+transformers==4.30.0
+torch==2.0.1

examples/text_to_speech_sync.py

@@ -15,4 +15,4 @@
 # Submit request
 job = model.execute(text=input_text, service_provider=service_provider)
 
-print (f"\n\n\n You can access the audio file here: {job.audio_url}")
+print(f"\n\n\n You can access the audio file here: {job.audio_url}")

requires-install.txt

@@ -1,2 +1,3 @@
 requests==2.28.1
 addict==2.4.0
+truss==0.4.8

slashml/__init__.py

@@ -3,5 +3,6 @@
 from slashml.text_summarization import TextSummarization  # noqa: F401,E402
 from slashml.speech_to_text import SpeechToText  # noqa: F401,E402
 from slashml.text_to_speech import TextToSpeech  # noqa: F401,E402
+from slashml.model_deployment import ModelDeployment  # noqa: F401,E402
 
 __all__ = ["TextSummarization", "SpeechToText", "TextToSpeech"]

slashml/model_deployment.py

@@ -0,0 +1,61 @@
+import json
+import requests
+import time
+import tarfile
+import truss
+from enum import Enum
+from .utils import generateURL, baseUrl, generateHeaders, formatResponse, getTaskStatus
+
+
+import os
+
+
+class ModelDeployment:
+    _base_url = baseUrl("model-deployment", "v1")
+    _headers = None
+
+    def __init__(self, api_key: str = None):
+        if api_key==None:
+            raise Exception("API Key is required for model deployment")
+        self._headers = generateHeaders(api_key)
+
+    def create_tar_gz(self, *, folder_path, tar_gz_filename):
+        with tarfile.open(tar_gz_filename, "w:gz") as tar:
+            tar.add(folder_path, arcname=os.path.basename(folder_path))
+
+    def deploy(self, *, model_name:str, model: str):
+        """Submit job"""
+        truss.create(model, 'my_model')
+        self.create_tar_gz(folder_path='my_model', tar_gz_filename='my_model.tar.gz')
+
+        url = generateURL(self._base_url, "models")
+        files = [("model_file", ("my_model.tar.gz", open('my_model.tar.gz', "rb"), "application/octet-stream"))]
+
+        payload = {
+            "model_name": model_name,
+        }
+
+        import pdb
+        pdb.set_trace()
+
+        response = requests.post(url, headers=self._headers, data=payload, files=files)
+
+        return formatResponse(response)
+
+    def status(self, *, model_version_id: str):
+        """Check job status"""
+        url = generateURL(self._base_url, "models", model_version_id, "status")
+        response = requests.get(url, headers=self._headers)
+        return formatResponse(response)
+    
+    def predict(self, model_version_id: str, model_input:str):
+        """Check job status"""
+
+        payload = json.dumps({
+            "model_input": model_input
+        })
+
+        url = generateURL(self._base_url, "models", model_version_id, "predict")
+        self._headers['Content-Type'] = 'application/json'
+        response = requests.post(url, headers=self._headers, data=payload)
+        return formatResponse(response)