Results from Vanna (with training) are worst than asking directly in the same LLM used by Vanna without any training

[tifDev]

Describe the bug
Vanna shows a worst result than LLM:

In Vanna:

With below training data:

df_ddl = vn.run_sql("SELECT type, sql FROM sqlite_master WHERE sql is not null")

for ddl in df_ddl["sql"].to_list():
    vn.train(ddl=ddl)

vn.train(
    ddl="""
        CREATE TABLE [Genre]
        (
            [GenreId] INTEGER  NOT NULL,
            [Name] NVARCHAR(120),
            CONSTRAINT [PK_Genre] PRIMARY KEY  ([GenreId])
        );
"""
)

vn.train(
    ddl="""
        CREATE TABLE [Track]
        (
            [TrackId] INTEGER  NOT NULL,
            [Name] NVARCHAR(200)  NOT NULL,
            [AlbumId] INTEGER,
            [MediaTypeId] INTEGER  NOT NULL,
            [GenreId] INTEGER,
            [Composer] NVARCHAR(220),
            [Milliseconds] INTEGER  NOT NULL,
            [Bytes] INTEGER,
            [UnitPrice] NUMERIC(10,2)  NOT NULL,
            CONSTRAINT [PK_Track] PRIMARY KEY  ([TrackId]),
            FOREIGN KEY ([AlbumId]) REFERENCES [Album] ([AlbumId])
            ON DELETE NO ACTION ON UPDATE NO ACTION,
            FOREIGN KEY ([GenreId]) REFERENCES [Genre] ([GenreId])
            ON DELETE NO ACTION ON UPDATE NO ACTION,
            FOREIGN KEY ([MediaTypeId]) REFERENCES [MediaType] ([MediaTypeId])
            ON DELETE NO ACTION ON UPDATE NO ACTION
        );
"""
)

# Sometimes you may want to add documentation about your business terminology or definitions.
vn.train(
    documentation="This project is related with music and contains Albuns, Artists, Genres, etc..."
)


# You can also add SQL queries to your training data. This is useful if you have some queries already laying around. You can just copy and paste those from your editor to begin generating new SQL.
vn.train(sql="select * from Genre;")


vn.train(
    question="How many tracks has each genre?",
    sql="""
        SELECT
            g.Name AS Genre,
            COUNT(t.TrackId) AS TotalTracks
        FROM
            Genre g
        JOIN
            Track t ON g.GenreId = t.GenreId
        GROUP BY
            g.Name;
    """,
)

Directly in LLM

And without any previous info or training data like I did in vanna:

Important note: I'm using the same LLM model in both

To Reproduce
Just try in Vann and through the model directly the same question.

Expected behavior
A better answer.

Error logs/Screenshots
Already provided.

Additional context
Connection to database is working fine, find below result provided by me placing the exact correct query:

[zainhoda]

Can you display the prompt that is sent to the LLM?

If you run vn.generate_sql it will show the log.