authentication fails when provided REPLICATE_API_TOKEN on versions 0.14.0 and greater

[Mohsin-Ul-Islam]

on versions 0.14.0 and greater, the client gives you have not passed an authentication token.

[extrange]

Having the same issue with 0.15.3.

I think it's due to the Client reading the REPLICATE_API_TOKEN on library import:

https://github.com/replicate/replicate-python/blob/30e4be55f96768b9d607cc8ddddb61f5aabcb3b3/replicate/client.py#L29C18-L40

I fixed this by importing the library after the environment variables are available, e.g.

# This needs to be before the llm library imports, otherwise they won't see the loaded environment variables
load_dotenv()

import replicate

[thijssdaniels]

I'm experiencing the same problem too. Using the default Colab available for fine-tuning a model on my own pictures

[mattt]

@Mohsin-Ul-Islam Thanks for reporting this. I apologize for any inconvenience there.

@extrange is exactly right that this new behavior is caused by a change in how the default client is initialized at the time of import.

@thijssdaniels The Colab notebook linked from the README demonstrates how to get started with Replicate.

Some users are reporting problems with this change, so I'm looking at different ways to communicate or mitigate this problem.

[Lazymindz]

@thijssdaniels i modified the notebook based on @extrange's suggestion and it now works.

You could refer to:
https://colab.research.google.com/drive/1D1MvippwX-qDm6MdUmuTJlnmCX3yIvUo?usp=sharing

changes:
Comment out the import of replicate at the start and import only after the env. variables are set.

@mattt just fyi. the README colab just state this either. Someone running the readme colab would still face this issue.

[mattt]

I just merged #170 with a change that should restore the previous behavior. Please upgrade to 0.15.4 and let me know if you're still seeing this issue. Thanks!

[Mhs467]

from dotenv import load_dotenv
import os
load_dotenv()

# Load environment variables from .env file
print("Loading environment variables...")

import replicate
api_token = os.getenv("REPLICATE_API_TOKEN")
print(f"API Token Loaded: {api_token}")

# Initialize the Replicate client with your API token
print("Initializing Replicate client...")
replicate_client = replicate.Client(api_token=api_token)

def predict_image(filename):
    try:
        print(f"Predicting image for file: {filename}")

        # Retrieve the model
        print("Retrieving the model...")
        model = replicate_client.models.get("tencentarc/gfpgan")
        print(f"Model retrieved: {model}")

        # List all available versions
        print("Listing all available versions...")
        for version in model.versions.list():
            print(f"Version ID: {version.id}, Created At: {version.created_at}")

        # Retrieve the latest version (or specify a particular one)
        print("Retrieving the latest version of the model...")
        latest_version = model.versions.get("ae80bbe1adce7d616b8a96ba88a91d3556838d4f2f4da76327638b8e95ea4694")  # You might replace "latest" with an actual version ID if needed
        print(f"Latest version retrieved: {latest_version}")

        # Prepare the inputs for the model prediction
        print("Preparing inputs for the model prediction...")
        inputs = {
            "img": open(filename, "rb"),  # Opening the image file in binary mode
            "scale": 2  # Rescaling factor
        }
        print(f"Inputs prepared: {inputs}")

        # Run the model prediction with the latest version
        print("Running the model prediction...")
        output = replicate.run("tencentarc/gfpgan:0fbacf7afc6c144e5be9767cff80f25aff23e52b0708f17e20f9879b2f21516c", input=inputs)
        print(f"Prediction output: {output}")

        return output  # Assuming the output is the URL of the restored image

    except Exception as e:
        print(f"An error occurred: {str(e)}")
        raise

# Example usage
# output_url = predict_image("path_to_your_image.jpg")
# print(f"Restored image URL: {output_url}")

I am using this code but I still get the same error. Can anyone help please? I have tried everything told above. It gives the version list pretty nicely but when I use the model thats latest it gives the error discussed above.

[mattt]

@Mhs467 In your code sample, the line starting output = replicate.run("tencentarc/gfpgan... uses the default Replicate client instance instead of the replicate_client constructed earlier. I don't see why the default client wouldn't pick up the API token from the environment automatically, but that at least accounts for the unexpected error.

[Mhs467]

Thank you for the speedy response matt. Means a lot. Let me implement your suggestion

…

On Thu, 22 Aug 2024, 15:24 Mattt, ***@***.***> wrote: @Mhs467 <https://github.com/Mhs467> In your code sample, the line starting output = replicate.run("tencentarc/gfpgan... uses the default Replicate client instance instead of the replicate_client constructed earlier. I don't see why the default client wouldn't pick up the API token from the environment automatically, but that at least accounts for the unexpected error. — Reply to this email directly, view it on GitHub <#169 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A6BGMXJ2SAL34L2PH5RTCJDZSW35JAVCNFSM6AAAAABM5YIPF6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBUGMYTANJRGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Mhs467]

import os from dotenv import load_dotenv load_dotenv() import replicate # Load environment variables from .env file # Initialize the Replicate client api_token = os.getenv("REPLICATE_API_TOKEN") replicate_client = replicate.Client(api_token=api_token) def predict_image(filename): try: print(f"Predicting image for file: {filename}") # Retrieve the model model = replicate_client.models.get("tencentarc/gfpgan") print(f"Model retrieved: {model}") # Retrieve the latest version latest_version = model.versions.get( "0fbacf7afc6c144e5be9767cff80f25aff23e52b0708f17e20f9879b2f21516c") print(f"Latest version retrieved: {latest_version}") # Prepare the inputs for the model prediction inputs = { "img": open(filename, "rb"), "scale": 2 } # Run the model prediction using the replicate_client instance output = replicate_client.run( "tencentarc/gfpgan:0fbacf7afc6c144e5be9767cff80f25aff23e52b0708f17e20f9879b2f21516c", input=inputs) print(f"Prediction output: {output}") return output except replicate.exceptions.ReplicateError as e: print(f"ReplicateError occurred: {e}") except Exception as e: print(f"An error occurred: {str(e)}") raise title: Invalid version or not permitted status: 422 detail: The specified version does not exist (or perhaps you don't have permission to use it?) Still the same error On Thu, Aug 22, 2024 at 3:35 PM Muhammad Hassan Saleem < ***@***.***> wrote:

…

Thank you for the speedy response matt. Means a lot. Let me implement your suggestion On Thu, 22 Aug 2024, 15:24 Mattt, ***@***.***> wrote: > @Mhs467 <https://github.com/Mhs467> In your code sample, the line > starting output = replicate.run("tencentarc/gfpgan... uses the default > Replicate client instance instead of the replicate_client constructed > earlier. I don't see why the default client wouldn't pick up the API token > from the environment automatically, but that at least accounts for the > unexpected error. > > — > Reply to this email directly, view it on GitHub > <#169 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/A6BGMXJ2SAL34L2PH5RTCJDZSW35JAVCNFSM6AAAAABM5YIPF6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBUGMYTANJRGY> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >

[mattt]

@Mhs467 Hmm, are you using the latest version of the replicate package? What do you see in pip list?

[Mhs467]

replicate 0.31.0 This one. When this didnt work I tried downgrading 0.15.4 but that didn't work to so I reverted back and tried this again. Same error. Let me ask a question and that might be a little silly too: Is payment method required before we can use the API or can we use it for free for some limited amount of requests.

…

On Thu, Aug 22, 2024 at 4:16 PM Mattt ***@***.***> wrote: @Mhs467 <https://github.com/Mhs467> Hmm, are you using the latest version of the replicate package? What do you see in pip list? — Reply to this email directly, view it on GitHub <#169 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A6BGMXINQJGQ45IIDEYUEMDZSXB7BAVCNFSM6AAAAABM5YIPF6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBUGQYTCMJXHA> . You are receiving this because you were mentioned.Message ID: ***@***.***>