Configured providers but twinny not sending any requests to provider.

[Duoquote]

Describe the bug
I have setup the following providers and I checked with curl that /api/generate endpoint on http://duodesk.duo:11434 works, the extension shows loading circle but is not sending any requests. Also tried setting Ollama Hostname setting to duodesk.duo, but no luck.

To Reproduce
Just added the providers I have attached.

Expected behavior
Should work with the providers I have I think?

Screenshots

Logging
Logging is enabled but not sure where am I supposed to see the logs, checked Output tab but there is no entry for twinny.

API Provider
Ollama running at http://duodesk.duo:11434 in local network.

Chat or Auto Complete?
Both

Model Name
codellama:7b-code

Desktop (please complete the following information):

• OS: Windows
• Version: 11

Additional context

[Duoquote]

I was looking at other issues and saw where I could see the logs, the log is as follows:

Looks like hostname is set incorrectly, how do I change it?

[rjmacarthy]

Maybe try a restart? The settings look correct to me. Also in the extension settings change the Ollama settings too. Click the cog in the extension header, there are some api settings for ollama in there too.

[randomeduc]

How can I check some logs? Because I have the same problem on Windows, I'm running VS Code from WSL:Ubuntu

1. Install Ollama and check if is running at http://localhost:11434

2. Install the models
   

3. Install and configure the vscode extension
   Settings
   

Providers

4. Test the chat
   

Got this message from vs code dev tool

But the loading spinner keep there, any help? Also try 127.0.0.1 as host but same results.

[jpaveg]

There is an issue with Twinny and WSL connected VSCode windows.

Continue extension works, but I can't get Twinny to work. Let me know if there's a way for me to help debug this.

I've also tried setting the host value to 127.0.0.1, localhost, and the default 0.0.0.0. Restarted Ollama, VSCode, my computer, but Twinny does not even reach the ollama server. I do not have this issue when I run Twinny from within a VSCode instance that is running without any remote container connectivity (in my case WSL2 specifically).

Relevant console logs(?):

ERR [Extension Host] Fetch error: TypeError: fetch failed
	at node:internal/deps/undici/undici:12345:11
	at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
	at async t.streamResponse (/home/user/.vscode-server/extensions/rjmacarthy.twinny-3.11.45/out/index.js:2:138539)
console.ts:137 [Extension Host] Fetch error: TypeError: fetch failed
	at node:internal/deps/undici/undici:12345:11
	at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
	at async t.streamResponse (/home/user/.vscode-server/extensions/rjmacarthy.twinny-3.11.45/out/index.js:2:138539)
y @ console.ts:137

[changty]

I have Twinny 3.12.0 on Linux. I'm running ollama with mistral-nemo and stable-code.
Curling the ollama works fine but Twinny just keeps spinning the circle, like in the screenshots above.
My project is running in a docker container, but that should not affect Twinny.

I do get a similar error message as above.

ERR [Extension Host] Fetch error: TypeError: fetch failed at node:internal/deps/undici/undici:12345:11 at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at async streamResponse (/root/.vscode-server/extensions/rjmacarthy.twinny-3.12.0/out/index.js:20901:22)

Any suggestions on how to fix this?

[unclemusclez]

i also just noticed this, i am using vscodium.

tbh i dont think this was always the case so i will update when i figure it out, hopefully.

[unclemusclez]

https://stackoverflow.com/questions/70590704/fetching-in-a-vscode-extension-node-fetch-and-nodehttp-issues

Judging from these two issues:

Cannot find module 'node:http` on AWS Lambda v14 and
Problem with importing node-fetch v3.0.0 and node v16.5.0

it looks like the upgrade from node-fetch v3 to v3.1 was "troublesome" and resulted in the error you are seeing for some.

A few users are downgrading to v2 (or you might try v3 rather than v3.1 which you are using).

Uninstall node-fetch and npm install -s node-fetch@2

[pacman100]

Facing the same issue. Any updates?

[disarticulate]

localhost and 0.0.0.0 not working for me. Have a responding ollama on both my local machine and remote.

UI responds "==## ERROR ##== : fetch failed"

In devtools, no requests are made, so I assume its on the backend.

The last error says:

ERR [Extension Host] �[91m [ERROR_twinny] �[32m Message: Fetch error 
 �[91m Error Type: TypeError 
  Error Message: fetch failed 
 �[31m TypeError: fetch failed
	at node:internal/deps/undici/undici:13392:13
	at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
	at async globalThis.fetch (file:///home/cy/.vscode-server/bin/138f619c86f1199955d53b4166bef66ef252935c/out/vs/workbench/api/node/extensionHostProcess.js:159:19800)
	at async streamResponse (/home/cy/.vscode-server/extensions/rjmacarthy.twinny-3.19.23-linux-x64/out/index.js:58388:22)
console.ts:137 [Extension Host]  [ERROR_twinny]  Message: Fetch error 

error goes here:

twinny/src/extension/api.ts

Line 103 in d528d53

log.logConsoleError(Logger.ErrorType.Fetch_Error, "Fetch error", error)

The fetch config from the stream response:

[Extension Host] [twinny] ***Twinny Stream Debug***
    Streaming response from 0.0.0.0:11434.
    Request body:
    {
  "model": "llama3.1:latest",
  "stream": true,
  "messages": [
    {
      "role": "user",
      "content": "yhello"
    },
    {
      "role": "assistant",
      "content": "==## ERROR ##== : fetch failed"
    },
    {
      "role": "user",
      "content": "\n  Generate a title for this conversation in under 10 words.\n  It should not contain any special characters or quotes.\n"
    }
  ],
  "keep_alive": "5m",
  "options": {
    "temperature": 0.2,
    "num_predict": 2048
  }
}

    Request options:
    {
  "hostname": "0.0.0.0",
  "port": 11434,
  "path": "/api/chat",
  "protocol": "http",
  "method": "POST",
  "headers": {
    "Content-Type": "application/json",
    "Authorization": "Bearer undefined"
  }
}

I can't see in ollama API but "Authorization" header should be excluded if it's undefined. Looks like there's several files setting Authorization to "Bearer undefined".

Added a pull request here: #422 there's several other files using the fetch.

I don't think it's the actual error though.

[disarticulate]

I just noticed though, my project is using SSH, and this plugin is being installed on the SSH vscode server, which isn't going to have access. That's definitely an error. It should probably be installed locally unless otherwise indicated.

[disarticulate]

I just noticed though, my project is using SSH, and this plugin is being installed on the SSH vscode server, which isn't going to have access. That's definitely an error. It should probably be installed locally unless otherwise indicated.

This is definitely a problem. I opened up a local workspace and it connects. I'm going to open a specific issue.

[unclemusclez]

localhost and 0.0.0.0 not working for me. Have a responding ollama on both my local machine and remote.

UI responds "==## ERROR ##== : fetch failed"

In devtools, no requests are made, so I assume its on the backend.

The last error says:

ERR [Extension Host] �[91m [ERROR_twinny] �[32m Message: Fetch error 
 �[91m Error Type: TypeError 
  Error Message: fetch failed 
 �[31m TypeError: fetch failed
	at node:internal/deps/undici/undici:13392:13
	at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
	at async globalThis.fetch (file:///home/cy/.vscode-server/bin/138f619c86f1199955d53b4166bef66ef252935c/out/vs/workbench/api/node/extensionHostProcess.js:159:19800)
	at async streamResponse (/home/cy/.vscode-server/extensions/rjmacarthy.twinny-3.19.23-linux-x64/out/index.js:58388:22)
console.ts:137 [Extension Host]  [ERROR_twinny]  Message: Fetch error 

error goes here:

twinny/src/extension/api.ts

Line 103 in d528d53

log.logConsoleError(Logger.ErrorType.Fetch_Error, "Fetch error", error)

The fetch config from the stream response:

[Extension Host] [twinny] ***Twinny Stream Debug***
    Streaming response from 0.0.0.0:11434.
    Request body:
    {
  "model": "llama3.1:latest",
  "stream": true,
  "messages": [
    {
      "role": "user",
      "content": "yhello"
    },
    {
      "role": "assistant",
      "content": "==## ERROR ##== : fetch failed"
    },
    {
      "role": "user",
      "content": "\n  Generate a title for this conversation in under 10 words.\n  It should not contain any special characters or quotes.\n"
    }
  ],
  "keep_alive": "5m",
  "options": {
    "temperature": 0.2,
    "num_predict": 2048
  }
}

    Request options:
    {
  "hostname": "0.0.0.0",
  "port": 11434,
  "path": "/api/chat",
  "protocol": "http",
  "method": "POST",
  "headers": {
    "Content-Type": "application/json",
    "Authorization": "Bearer undefined"
  }
}

I can't see in ollama API but "Authorization" header should be excluded if it's undefined. Looks like there's several files setting Authorization to "Bearer undefined".

Added a pull request here: #422 there's several other files using the fetch.

I don't think it's the actual error though.

are you using the ollama template with an openai endpoint, or vice-versa?

[rjmacarthy]

Please let me know if we should re-open, thanks!

[disarticulate]

are you using the ollama template with an openai endpoint, or vice-versa?

No. I'm fairly certain my issue is related to the plugin installing in the SSH workspace which doesn't have access to my localhost run Ollama. see #423 for what the most likely cause is of multiple error reports.

The work around at the moment for SSH workplace is to provide a IP and open port to the ollama server if you're working on remote; or use a local code source. It might be possible in the open ports section, but I don't know if that's two-way; I think it's just a one way, but don't know how the SSH workspace plugin is establishing the connection.