support api_endpoint for ollama

[luixiao0]

No description provided.

[SilasMarvin]

This is really awesome thank you!

What do you think about explicitly adding a chat_endpoint and a generate_endpoint in the config options for Ollama instead of just endpoint? We do this for every other API backend. If a user uses Ollama remotely are we sure they will have chat at /api/chat and generate at /api/generate?

[luixiao0]

This is really awesome thank you!

What do you think about explicitly adding a chat_endpoint and a generate_endpoint in the config options for Ollama instead of just endpoint? We do this for every other API backend. If a user uses Ollama remotely are we sure they will have chat at /api/chat and generate at /api/generate?

I think Adding separate chat_endpoint and generate_endpoint for Ollama isn't necessary for now because once Ollama is running, it provides both chat and generate functionalities at /api/chat and /api/generate. But it is Fine to have 2 config options, providing better consistency across providers

[SilasMarvin]

If you can make the suggested changes and resolve the merge conflict I will merge it in! Thank you!

[luixiao0]

resolved, 3a351ff237f4cf

[SilasMarvin]

resolved, 3a351ff237f4cf

Awesome thank you. If you can implement the changes I suggested above I will merge it

[luixiao0]

Sorry, I misunderstood your suggestion. The suggested changes have been committed.

[SilasMarvin]

Sorry, I misunderstood your suggestion. The suggested changes have been committed.

Don't apologize you are totally fine. I think we are having some issues communicating. If you scroll to the top of the page, you will see I have added some suggestions for changes. You can click commit on those directly, or you can add them yourself. Please verify they work and I will merge the PR. Thank you for your contributions!

[luixiao0]

checked, work at my env, ican set up with this config

generate example
{
    "lsp-ai.serverConfiguration": {
        "memory": {
            "file_store": {}
          },
          "models": {
            "model1": {
              "type": "ollama",
              "completions_endpoint": "http://remote:11434/api/generate",
              "model": "deepseek-coder",
            }
          }
    },
}

chat example
{
    "lsp-ai.serverConfiguration": {
        "memory": {
            "file_store": {}
          },
          "models": {
            "model1": {
              "type": "ollama",
              "chat_endpoint": "http://localhost:11434/api/chat",
              "model": "deepseek-coder",
            }
          }
    },
    "lsp-ai.generationConfiguration": {
      "model": "model1",
      "parameters": {
        "messages": [
          {
            "role": "system",
            "content": "Instructions:"
          }
        ]
      }
    },
}

i didn't find comment in "Files changed" tab, are you talking about "explicitly adding a chat_endpoint and a generate_endpoint in the config options for Ollama"?
Do you mean when only chat_endpoint is set, lsp-ai should use chat completions instead of generate completions?
Both cases

[SilasMarvin]

Thanks for testing it locally. These are the suggestions I am talking about:

We shouldn't call it completions_endpoint as Ollama calls it generate. I liked the defaults you had before.

If you can commit those suggestions and test it, I will merge it!

[luixiao0]

Is github not working as intended? i cannot see them on my side, nor my mailbox, i will make the change shown in your screenshot

I liked the defaults you had before.

i will revert them at next commit

suggested by gpt-4o: try "submitreview"

[SilasMarvin]

Is github not working as intended? i cannot see them on my side, nor my mailbox, i will make the change shown in your screenshot

I liked the defaults you had before.

i will revert them at next commit

suggested by gpt-4o: try "submitreview"

Oh super strange. I have attached the two other suggestions that I made.

[luixiao0]

Is github not working as intended? i cannot see them on my side, nor my mailbox, i will make the change shown in your screenshot

I liked the defaults you had before.

i will revert them at next commit
suggested by gpt-4o: try "submitreview"

Oh super strange. I have attached the two other suggestions that I made.

I've addressed the suggested modifications and it's ready for your feedback

[SilasMarvin]

Looks pretty good!

You should be able to change:

.post(self
                .configuration
                .generate_endpoint
                .as_ref()
                .unwrap_or(&"http://localhost:11434/api/generate".to_string())

To:

.post(self.configuration.generate_endpoint.as_deref().unwrap_or("http://localhost:11434/api/generate"))

[luixiao0]

my bad, it's done, never heard of deref before

[SilasMarvin]

No worries! Awesome work, I'm going to test locally and merge it tonight!

Thanks for contributing!