Try NLU Model Error

gustavocami · 3 weeks ago

Hello!

I was creating new Vocabulary Sources and Entities for one NLU Model (it's included in a Multi-lingual NLU Model set).

After creation, I trained the model, and then when I tried to test in the NLU Workbench, the website threw an error:

Exception caught in the predict call - reason: {"status":"failure","exception":"sn_ml.MLServiceUtil.predict(): com.snc.ml.prediction.common.ServiceException: java.lang.Exception: SE0217:The solution metadata is null. Unable to retrieve: ml_x_bueu_global_global_ab2768e31b5435907c4bdd39cd4bcba9_1 [knowledge_pb.knowledge_pbCompiledFsa.json, knowledge_pb.knowledge_pb.zip, lookupSourceConfig.json]"}

I tried to Sync the Vocabulary sources again and retrain, but it didn't work.

Vikram Reddy · 3 weeks ago

Hello @gustavocami,

I'm assuming the new Entity you added is backed by a vocabulary source of type Static Lookup, since that knowledge_pb bundle in the error (lookupSourceConfig.json, knowledge_pbCompiledFsa.json) is what a lookup-type fuzzy matcher solution generates, separate from the intent model itself. A Sync alone won't fix it if the underlying solution reference is stale, so before retraining again, check:

the Type field on the new record in sys_nlu_vocabulary, confirm it really is Static Lookup
the solution_name field on that record, it's likely still pointing at ml_x_bueu_global_global_ab2768e31b5435907c4bdd39cd4bcba9_1
whether that sys_id actually has a matching row in ml_solution, a missing or empty one is your null metadata
clear out solution_name on the vocabulary record, then retrain so a fresh solution gets generated instead of reusing the broken one

Thank you,
Vikram Karety
Octigo Solutions INC

gustavocami · 3 weeks ago

The solution_name ml_x_bueu_global_global_ab2768e31b5435907c4bdd39cd4bcba9_1 and ml_x_bueu_global_global_43645831c3a1bad0121233ff0501315e_1 indicated in the error message don't exist. I tried to find in the tables of vocabulary sources, ml_solution, and ml_model_artifact, but none of them had these solution names. Probably they don't exist anymore.

Vikram Reddy · 3 weeks ago

Hi @gustavocami,

Clear the solution_name field on whatever vocabulary source is pointing at those two IDs, then retrain. That's the fix once a referenced solution has genuinely been removed from ml_solution.

Those ml_x_bueu_global_global_... names are Fuzzy Matcher solutions generated off a vocabulary source, not the NLU model's own solution record. The reference to them lives on the sys_nlu_vocabulary record itself, in the solution_name field, not inside ml_solution or ml_model_artifact, so searching those two tables for the string won't turn up anything even in a healthy setup. When the ID is gone everywhere, it's usually one of two causes:

Update set import: the vocabulary source came in from another instance via update set, but the underlying ml_solution and ml_model_artifact records were never regenerated locally, so solution_name still holds a stale reference from the source environment.
Duplicate cleanup: ml_solution ended up with two records for the same solution and version (one active=false), and the inactive one that got deleted or purged happened to be the one this vocabulary source pointed at.

To fix it:

Open the vocabulary sources in sys_nlu_vocabulary that are Type = Static Lookup or Table, and check solution_name on each against the two IDs from your error.
Clear solution_name on the matching record(s).
Retrain the NLU model. Training regenerates the Fuzzy Matcher solution, writes a fresh solution_name back to the vocabulary source, and creates new ml_solution and ml_model_artifact records.
Run Try It again once training finishes.

If neither vocabulary source actually shows those IDs, check ml_solution sorted by Updated for duplicate rows under the same solution name and version. That mismatch between an active and a stale inactive copy is a known pattern that also blocks publish and test.

Post the vocabulary source type and whether this model came in through an update set if it's still stuck, happy to dig further.

Thank you,
Vikram Karety
Octigo Solutions INC

gustavocami · 3 weeks ago

Hi, @Brian Bakker and @Vikram Reddy .

I also searched the sys_nlu_vocabulary table for the solution name indicated on the error message. It doesn't exist.

These vocabulary sources did not come from an Update Set, I'm developing in the DEV instance.

Below are the vocabulary sources indicated in the error message. I removed their solution_name values and retrained (it didn't work).

The error log got bigger than the last time, now it displays this: