How to reproduce the results on SWE-Bench

[mannaandpoem]

How to repreduce this result on SWE-Bench: https://xwang.dev/blog/2024/opendevin-codeact-1.0-swebench/

I'm using the main branch, selecting CodeActAgent as the agent and gpt-4-1106-preview as the base model. Given a link to an issue, it can be fixed to a certain extent, but it does not actually modify the local file, and it does not execute the git commit and git push commands.

[prithvi151080]

@netgfx what is the ollama URL that you are using...if its running in docker plz use host.docker.internal:11434 as the URL. hope this helps

[netgfx]

I tried that but didn't work, also not sure how it would work, meaning that it is hosted on a separate server (with docker yes) but different domain altogether

[prithvi151080]

@netgfx find the IP address of the docker container (run docker inspect <container ID> and try to ping that. If ur able to ping that container, then use the IP as the URL with the port 11434 and it will work. if ping fails, then there is some network issue. check the subnets of the machine from where ur trying to ping and the container of ollama. if they are diff subnets then use NAT translation so that ur able to ping. hope this helps.

[HenryHengZJ]

@netgfx can you try if changing the url to http://host.docker.internal:11434?

[netgfx]

@HenryHengZJ I did try it but nothing, still get the fetch failed error

[bigsk1]

yep this solved my issue, ollama on wsl and flowise in container bridge network, only by using http://host.docker.internal:11434 would it work!

[netgfx]

I tried the host.docker.internal:11434 and even went and setup both images from my local environment (WSL Ubuntu) But it still fails, this is my setup

I can ping host.docker.internal:11434 from the docker cli, and I can also access it from the browser, it seems only flowise isn't able to connect with it for some reason :(

[Z3n7r4ck3r]

It also fails with a non-dockerized instance of ollama and a dockerized instance of flowise, both on the same windows machine. Moreover, on the same machine a dockerized instance of open-webui is perfectly fetching models served by ollama.

found solution here https://github.com/ollama/ollama/blob/main/docs/faq.md

It works fine now.