When using Ollama, a long loop of Initializing Ollama Model happens before any output

[FGhrawi]

Describe the bug

When using Ollama, the model being used is loaded over and over in a loop (sometimes 10-15+ times) in the terminal before any output or agent decision-making happens.
Fwiw, one model is configured at a time in .env (so no swapping is being done)

To Reproduce

Use any model with the Ollama provider

Expected behavior

Model loads once. (Since ollama is on keepalive and only 1 model is being used)

Screenshots

Additional context
I am on a fresh clone of the repo with a basic character configuration.
Fwiw, I have 24gb vram, and this happens even on smaller models.

[drew-royster]

@FGhrawi can you pull latest and confirm that this is fixed?

[yodamaster726]

I did a fresh clone on the latest code and the code keeps wanting to download and run the local_llama.
I have the OLLAMA_MODEL set in my .env. Here is what have:

#Set to Use for New OLLAMA provider
OLLAMA_SERVER_URL= #Leave blank for default localhost:11434
OLLAMA_MODEL=hermes3
OLLAMA_EMBEDDING_MODEL= #default mxbai-embed-large
#To use custom model types for different tasks set these
SMALL_OLLAMA_MODEL= #default llama3.2
MEDIUM_OLLAMA_MODEL= #default hermes3
LARGE_OLLAMA_MODEL= #default hermes3:70b

This is what I have in my defaultCharacter.

modelProvider: ModelProviderName.OLLAMA,
settings: {
    secrets: {},
    voice: {
        model: "en_US-hfc_female-medium",
    },
    embeddingModel: "mxbai-embed-large"
},

Code is also tries to load the wrong OLLAMA embedding model too by default - hence having to set it here.

I'm debugging the problem now and working on a fix for it.

[lakshya404stc]

Is this issue still valid??

[yodamaster726]

yes, ollama and llama local got merged together and the ollama logic is not working right. getting close to a fix.
example, its downloading the local llama instead of using ollama if it's configured

[drew-royster]

@yodamaster726 what is your character file like? I had ollama as the model in my character and it seemed to use ollama just fine.

[yodamaster726]

I tried the latest tag v0.1.3 and then tried the latest code from yesterday. My character file was the default one.

updates to fix this problem: #521

[dr-fusion]

just tested with the latest code and used ModelProviderName.LLAMALOCAL,

Seems like its still going in a loop.

[MERNinja]

ModelProviderName.LLAMALOCAL is still in loop with alpha.1 tag release. Anyone found a solution?

[AIFlowML]

ModelProviderName.LLAMALOCAL is still in loop with alpha.1 tag release. Anyone found a solution?

Did you try last version we just release ?
If is not working at all i wil look in the code.
Please let me know.

[Luucky]

I tried the following fix locally which prevented the looping on LLAMA_LOCAL model provider
https://github.com/elizaOS/eliza/pull/1755/files