Created
December 8, 2022 07:36
-
-
Save kasperjunge/4ec27724272109b9a413e4d1477ef7cd to your computer and use it in GitHub Desktop.
ChatGPT answer to: Implement me a hyper-optimized service for serving predictions by transformer models
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from fastapi import FastAPI | |
import transformers | |
import torch | |
app = FastAPI() | |
# Load the tokenizer and model | |
tokenizer = transformers.AutoTokenizer.from_pretrained("model_name") | |
model = transformers.AutoModelForTokenClassification.from_pretrained("model_name") | |
# Use torch.jit.trace to create a torch.jit.ScriptModule from the model | |
traced_model = torch.jit.trace(model, example_inputs=inputs) | |
# Check if a CUDA-enabled GPU is available | |
if torch.cuda.is_available(): | |
# If a GPU is available, move the model and input tensors to the GPU | |
traced_model = traced_model.to("cuda") | |
inputs = inputs.to("cuda") | |
@app.post("/predict") | |
def predict(input: str): | |
# Tokenize the input | |
tokens = tokenizer.encode(input) | |
# Convert the tokens to their numeric IDs | |
input_ids = tokenizer.convert_tokens_to_ids(tokens) | |
# Create a tensor with the input IDs | |
inputs = torch.tensor([input_ids]) | |
# If a GPU is available, move the input tensor to the GPU | |
if torch.cuda.is_available(): | |
inputs = inputs.to("cuda") | |
# Prevent PyTorch from tracking the gradients of the model's weights | |
with torch.no_grad(): | |
# Use the traced model to generate predictions | |
predictions = traced_model(inputs) | |
# Get the first prediction | |
first_prediction = predictions[0] | |
# If a GPU is available, move the prediction tensor back to the CPU | |
if torch.cuda.is_available(): | |
first_prediction = first_prediction.to("cpu") | |
# Convert the prediction tensor to a list of predicted labels | |
predicted_labels = [tokenizer.decode(g) for g in torch.argmax(first_prediction, axis=1)] | |
# Return the predicted labels as the response | |
return {"labels": predicted_labels} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment