Nota: También he probado creando una clase con un código de python simple y da el mismo error. ¿Puede ser que esta imagen de docker tenga problemas con Python?

Class St.Vectorsearch.Vector Extends %RegisteredObject
{
ClassMethod Embedding(Test) [ Language = python ]
{
   return (Test)
}
}

Compilation started on 04/16/2026 06:12:25 with qualifiers 'cuk'
Compiling class St.Vectorsearch.Vector
ERROR #7802: Worker job/s '1199:40' unexpectedly shut down in group '#Default:(2988722416963):0'.
ERROR #7812: Work queue unexpectedly removed, shutting down.
ERROR #5002: ObjectScript error: <THROW>WaitForComplete+215^%SYS.WorkQueueMgr *%Exception.StatusException ERROR #7802: Worker job/s '1199:40' unexpectedly shut down in group '#Default:(2988722416963):0'.
ERROR #7812: Work queue unexpectedly removed, shutting down.
Detected 3 errors during compilation in 1.014s.

Hola @Luis Angel Pérez Ramos 

He reiniciado mi docker para que vuelva a compilar todo desde cero para ver si así volvía a la vida, pero ahora me aparece el siguiente error cuando compila la clase St.Vectorsearch.Vector

USER>do $system.OBJ.Compile("St.Vectorsearch.Vector","ck")
Compilation started on 04/16/2026 06:05:04 with qualifiers 'ck'
Compiling class St.Vectorsearch.Vector
ERROR #7802: Worker job/s '878:23' unexpectedly shut down in group '#Default:(4743230725894):0'.
ERROR #7812: Work queue unexpectedly removed, shutting down.
ERROR #5002: ObjectScript error: <THROW>WaitForComplete+215^%SYS.WorkQueueMgr *%Exception.StatusException ERROR #7802: Worker job/s '878:23' unexpectedly shut down in group '#Default:(4743230725894):0'.
ERROR #7812: Work queue unexpectedly removed, shutting down.
Detected 3 errors during compilation in 1.003s.

He probado a usar tanto la versión de sentence-transforme que me habías pasado como usando la última versión, y da el mismo error al compilarlo.

¿Alguna idea?

Esta es mi clase St.Vectorsearch.Vector

Class St.Vectorsearch.Vector Extends %RegisteredObject
{
ClassMethod Embedding(Test) [ Language = python ]
{
   from transformers import AutoTokenizer, AutoModel
   import torch
   import torch.nn.functional as F
   #Mean Pooling - Take attention mask into account for correct averaging
   def mean_pooling(model_output, attention_mask):
       token_embeddings = model_output[0] #First element of model_output contains all token embeddings
       input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
       return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)

   # Sentences we want sentence embeddings for
   # sentences = ['This is an example sentence', 'Each sentence is converted']
   sentences = [Test]
   # Load model from HuggingFace Hub
   tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
   model = AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
   # Tokenize sentences
   encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
   # Compute token embeddings
   with torch.no_grad():
       model_output = model(**encoded_input)
   # Perform pooling
   sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
   # Normalize embeddings
   sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)
   #print(sentence_embeddings)
   return str(sentence_embeddings[0].tolist())
}
}

Estoy usando la imagen de docker de community

No, en ningún momento. Simplemente se queda esperando una respuesta del método Embedding y ahí se queda