AF
Ash Framework•2mo ago
Abu kumathra

ashai vectorize

There is an issue where the code keeps embedding vectors even though they were already embedded I am on postgres 16, 
 sudo -u postgres psql  -d t_db -c "select count(*) from poems;"
 count 
-------
  1437
(1 row)
 sudo -u postgres psql  -d t_db -c "select count(*) from poems;"
 count 
-------
  1437
(1 row)
 select count(id) from oban_jobs where queue = 'poem_vectorize';
 count 
-------
  4314
(1 row)
 select count(id) from oban_jobs where queue = 'poem_vectorize';
 count 
-------
  4314
(1 row)
No description
10 Replies
ZachDaniel
ZachDaniel•2mo ago
What does your trigger look like?
Abu kumathra
Abu kumathraOP•2mo ago
  vectorize do
    attributes title: :vectorized_title,
               meter: :vectorized_meter,
               # rhyme: :vectorized_rhyme,
               topics: :vectorized_topics

    # strategy :manual
    strategy :ash_oban
    ash_oban_trigger_name :vectorize
    embedding_model Taleed.OpenAIEmbeddingModel
  end

  oban do
    triggers do
      trigger :vectorize do
        action :ash_ai_update_embeddings
        worker_read_action :read
        worker_module_name __MODULE__.AshOban.Worker.UpdateEmbeddings
        scheduler_module_name __MODULE__.AshOban.Scheduler.UpdateEmbeddings
        # list_tenants MyApp.ListTenants
      end
    end
  end
  vectorize do
    attributes title: :vectorized_title,
               meter: :vectorized_meter,
               # rhyme: :vectorized_rhyme,
               topics: :vectorized_topics

    # strategy :manual
    strategy :ash_oban
    ash_oban_trigger_name :vectorize
    embedding_model Taleed.OpenAIEmbeddingModel
  end

  oban do
    triggers do
      trigger :vectorize do
        action :ash_ai_update_embeddings
        worker_read_action :read
        worker_module_name __MODULE__.AshOban.Worker.UpdateEmbeddings
        scheduler_module_name __MODULE__.AshOban.Scheduler.UpdateEmbeddings
        # list_tenants MyApp.ListTenants
      end
    end
  end
this is the trigger as you can see all the jobs were done in the vectorize queue for the poems
ZachDaniel
ZachDaniel•2mo ago
      trigger :vectorize do
        action :ash_ai_update_embeddings
        worker_read_action :read
        scheduler_cron nil # <- add this
        worker_module_name __MODULE__.AshOban.Worker.UpdateEmbeddings
      end
      trigger :vectorize do
        action :ash_ai_update_embeddings
        worker_read_action :read
        scheduler_cron nil # <- add this
        worker_module_name __MODULE__.AshOban.Worker.UpdateEmbeddings
      end
It runs it on a schedule by default which you don't want is it in our docs like that? If so plase open an issue or even bettr a PR
Abu kumathra
Abu kumathraOP•2mo ago
yes actually it does have scheduler_cron nil, but I think the compiler was complaining about something
Abu kumathra
Abu kumathraOP•2mo ago
it should be set to false instead
No description
Abu kumathra
Abu kumathraOP•2mo ago
Abu kumathra
Abu kumathraOP•2mo ago
also there is no mention of the need to add the oban queue to the config.exs the idea of the scheduler is great but it can be improved by actually batching the get embedding call [this is already been done when there is different attributes in the same] + only touching non embedded rows this would be a good feature, I am willing to help but I will need links to guides. also I am new to the framework
ZachDaniel
ZachDaniel•2mo ago
Yeah, we want to make ash_oban support bulk updates once we support like a batch_size and have it fetch n records and bulk update them then we can just use that in this action Can you PR the addition of those options to the docs?
Abu kumathra
Abu kumathraOP•2mo ago
as far as I remember open ai has a max token limit on the embedding request so maybe a way to handle that too would be great will do
Abu kumathra
Abu kumathraOP•2mo ago
https://github.com/ash-project/ash_ai/pull/109 done
GitHub
impr: Switching off the scheduler+Mention of the need to add the qu...
Contributor checklist Leave anything that you believe does not apply unchecked. I accept the AI Policy, or AI was not used in the creation of this PR. Bug fixes include regression tests Chores ...

Did you find this page helpful?