You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I would like to have a possibility to change default prompts (text, image, summarization) without recompiling the project.
Describe the benefits this would bring to existing Hoarder users
With this feature, users will be able to fully customize the prompts used to communicate with LLMs.
There are some open issues, where current prompts are tweaked to work better with users' local language models, but each model works differently - prompts which work perfectly with llama3.2 might not work with deepseek-r1 or qwen.
In my case, the default prompt didn't always return satisfying results, even when I created some tagging rules using Hoarder settings.
Can the goal of this request already be achieved via other means?
Yes, Hoarder can be rebuilt with custom prompt. It is very frustrating, though - in order to keep Hoarder version up to date, each new release must be rebuilt with prompt changes.
Have you searched for an existing open/closed issue?
I have searched for existing issues and none cover my fundamental request
Additional context
My suggestion would be to allow prompts customization through .env variables. This way the default prompts can be kept in the codebase and they could be updated on demand without rebuilding the application / container. Such approach would also be consistent with other inference configuration options - for example OpenAI / Ollama URLs, which are also global for the whole Hoarder instance.
I already have some code which implements these changes, I can open a PR if such solution is acceptable.
The text was updated successfully, but these errors were encountered:
Please have a look at my comment here: #1029 (comment)
This is intentionally restricted. I'm wiling to add more flexibility to customize pre-packaged rules, but I'm not planning to give up the control on the prompt structure itself to allow me to evolve it in the future.
Describe the feature you'd like
I would like to have a possibility to change default prompts (text, image, summarization) without recompiling the project.
Describe the benefits this would bring to existing Hoarder users
With this feature, users will be able to fully customize the prompts used to communicate with LLMs.
There are some open issues, where current prompts are tweaked to work better with users' local language models, but each model works differently - prompts which work perfectly with llama3.2 might not work with deepseek-r1 or qwen.
In my case, the default prompt didn't always return satisfying results, even when I created some tagging rules using Hoarder settings.
Can the goal of this request already be achieved via other means?
Yes, Hoarder can be rebuilt with custom prompt. It is very frustrating, though - in order to keep Hoarder version up to date, each new release must be rebuilt with prompt changes.
Have you searched for an existing open/closed issue?
Additional context
My suggestion would be to allow prompts customization through .env variables. This way the default prompts can be kept in the codebase and they could be updated on demand without rebuilding the application / container. Such approach would also be consistent with other inference configuration options - for example OpenAI / Ollama URLs, which are also global for the whole Hoarder instance.
I already have some code which implements these changes, I can open a PR if such solution is acceptable.
The text was updated successfully, but these errors were encountered: