ERROR: failed to solve: circular dependency detected on stage: weights
Hello,
Trying to push a model with --separate-weights fails with this error. Works fine when the flag is not used. I am using cog version 0.9.7. I also tried deleting .cog , removed the models/ folder from the auto-generated .dockerignore
Building Docker image from environment in cog.yaml as r8.im/xxxxx/xxxxx...
[+] Building 187.8s (8/8) FINISHED docker:desktop-linux
=> [internal] load .dockerignore 0.0s
=> => transferring context: 1.47kB 0.0s
=> [internal] load build definition from Dockerfile 0.0s
=> => transferring dockerfile: 141B 0.0s
=> resolve image config for docker.io/docker/dockerfile:1.4 1.1s
=> CACHED docker-image://docker.io/docker/dockerfile:1.4@sha256:9ba7531bd80fb0a858632727cf7a112fbfd19b17e94c4e84ced81e24ef1a0dbc 0.0s
=> [internal] load build context 98.2s
=> => transferring context: 10.62GB 98.2s
=> [1/1] COPY models /src/models 58.0s
=> preparing layers for inline cache 30.1s
=> exporting to image 0.0s
=> => exporting layers 0.0s
=> => writing image sha256:5be97ec87e2fb8b2f8f001848c916dd1a2031b3624ff2974d14e8558adc3c655 0.0s
=> => naming to r8.im/xxxxxx/xxxxxx 0.0s
[+] Building 3.7s (5/5) FINISHED docker:desktop-linux
=> [internal] load build definition from Dockerfile 0.0s
=> => transferring dockerfile: 2.47kB 0.0s
=> [internal] load .dockerignore 0.1s
=> => transferring context: 1.49kB 0.0s
=> resolve image config for docker.io/docker/dockerfile:1.4 3.4s
=> [auth] docker/dockerfile:pull token for registry-1.docker.io 0.0s
=> CACHED docker-image://docker.io/docker/dockerfile:1.4@sha256:9ba7531bd80fb0a858632727cf7a112fbfd19b17e94c4e84ced81e24ef1a0dbc 0.0s
Dockerfile:1
--------------------
1 | >>> #syntax=docker/dockerfile:1.4
2 | FROM python:3.11 as deps
3 | COPY .cog/tmp/build3627906443/cog-0.0.1.dev-py3-none-any.whl /tmp/cog-0.0.1.dev-py3-none-any.whl
--------------------
ERROR: failed to solve: circular dependency detected on stage: weights
ⅹ Failed to build runner Docker image: Failed to build Docker image: exit status 1
Below is cog.yaml
# Configuration for Cog ⚙️
# Reference: https://cog.run/yaml
build:
# set to true if your model requires a GPU
gpu: true
# a list of ubuntu apt packages to install
system_packages:
- "git"
# python version in the form '3.11' or '3.11.4'
python_version: "3.11"
python_requirements: requirements.txt
# predict.py defines how predictions are run on your model
predict: "predict.py:Predictor"
Same issue here with 0.9.7 and 0.9.8 on Linux.
I have the same problem on mac with cog version 0.9.8. Also try with the versions v0.9.9-alpha and v0.10.0-alpha7 and I only got new problems. Yesterday I was talking to Replicate technical support and they usually communicated the problem to the cog developers.
Hmm, seems like this isn't well maintained. No way to run custom models with everything baked into container without this :/ I guess this blocks any further evaluation of replicate as a provider
This has been the response I have had.
" This is a known huge friction point for pushing/building Cog models currently. I've flagged the issues with --separate-weights to the devs on the team that maintain Cog to prioritize a fix for it, but I won't have an ETA to share on when they'll be able to land that work.
I'm sorry about that. I know it's an impediment to the workflows for iterating on models. I promise we're thinking about how to solve it!
Thanks for your patience while we work out the kinks in the product."
What makes Replicate unfeasible for developing a model with heavy dependencies without a local GPU. The option remains to work with a rented server for development and upload the implementation to Replicate only when it is ready.
I still have this issue with cog 0.9.9 on Ubuntu. I have even docker clean'ed to a factory reset and can't fix it :(
@bfirsh ?
Duplicate of #1323
I fixed it in PR #1807. For anyone who can't wait you can download here.
@turian btw this isn't a dupe of #1323, it's similar but that one can be fixed by deleting .cog/ folder while this one can't.
@gurteshwar @hervenivon @Adrian-Serna @turian Sorry for the radio silence, folks. I dropped the ball on this, and I apologize for the inconvenience this caused.
I just opened a PR that should fix this regression here: https://github.com/replicate/cog/pull/1811. I haven't confirmed the fix from @LagPixelLOL's PR, but you're welcome to give that a try, too.
This is now available in Cog v0.9.13 (it will take a few hours for brew upgrade cog to work, but you can install from source or download a precompiled release from GitHub in the meantime). Thanks again for your patience. Please let me know if you're still having problems after upgrading.