4.6.2 ML_GENERATE_TABLE

The ML_GENERATE_TABLE routine runs multiple text generation or summarization queries in a batch, parallelly. The output generated for every input query is the same as the output generated by the ML_GENERATE routine.

This routine is available in HeatWave 9.0.1-u1 and later versions.

ML_GENERATE_TABLE Syntax

mysql> call sys.ML_GENERATE_TABLE('InputTableColumn', 'OutputTableColumn', [options]);
    
options: {
  JSON_OBJECT('key','value'[,'key','value'] ...)
    'key','value': {
    ['task', {'generation'|'summarization'}]
    ['model_id', {'mistral-7b-instruct-v1'|'llama2-7b-v1'|'llama3-8b-instruct-v1'}]
    ['context_column', 'ContextColumn']
    ['language', 'Language']
    ['temperature', Temperature]
    ['max_tokens', MaxTokens]
    ['top_k', K]
    ['top_p', P]
    ['repeat_penalty', RepeatPenalty]
    ['frequency_penalty', FrequencyPenalty]
    ['stop_sequences', JSON_ARRAY('StopSequence1'[,'StopSequence2'] ...)]
    ['batch_size', BatchSize]
    }
}

Following are ML_GENERATE_TABLE parameters:

• InputTableColumn: specifies the names of the input database, table, and column that contains the natural-language queries. The InputTableColumn is specified in the following format: DBName.TableName.ColumnName.

  • The specified input table can be an internal or external table.

  • The specified input table must already exist, must not be empty, and must have a primary key.

  • The input column must already exist and must contain text or varchar values.

  • The input column must not be a part of the primary key and must not have NULL values or empty strings.

  • There must be no backticks used in the DBName, TableName, or ColumnName and there must be no period used in the DBName or TableName.

• OutputTableColumn: specifies the names of the database, table, and column where the generated text-based response is stored. The OutputTableColumn is specified in the following format: DBName.TableName.ColumnName.

  • The specified output table must be an internal table.

  • If the specified output table already exists, then it must be the same as the input table. And, the specified output column must not already exist in the input table. A new JSON column is added to the table. External tables are read only. So if input table is an external table, then it cannot be used to store the output.

  • If the specified output table doesn't exist, then a new table is created. The new output table has key columns which contains the same primary key values as the input table and a JSON column that stores the generated text-based responses.

  • There must be no backticks used in the DBName, TableName, or ColumnName and there must be no period used in the DBName or TableName.

• options: specifies optional parameters as key-value pairs in JSON format. It can include the following parameters:

  • task: specifies the task expected from the large language model (LLM). Default value is generation. Possible values are:

    • generation: generates text-based content.

    • summarization: generates a summary for existing text-based content.

  • model_id: specifies the LLM to use for the task. Default value is mistral-7b-instruct-v1. Possible values are:

    • mistral-7b-instruct-v1

    • llama2-7b-v1

    • llama3-8b-instruct-v1

    To view the lists of supported LLMs, see LLMs.

  • context_column: specifies the table column that contains the context to be used for augmenting the queries and guiding the text generation of the LLM. The column must be specified in the following format: DBName.TableName.ColumnName. The specified column must be an existing column in the input table. Default value is NULL.

  • language: specifies the language to be used for writing queries, ingesting documents, and generating the output. To set the value of the language parameter, use the two-letter ISO 639-1 code for the language.

    Default value is en.

    For possible values, to view the list of supported languages, see Languages.

  • temperature: specifies a non-negative float that tunes the degree of randomness in generation. Lower temperatures mean less random generations.

    Default value is 0.

    Possible values are float values between 0 and 5.

    It is suggested that:

    • To generate the same output for a particular prompt every time you run it, set the temperature to 0.

    • To generate a random new statement for a particular prompt every time you run it, increase the temperature.

  • max_tokens: specifies the maximum number of tokens to predict per generation using an estimate of three tokens per word. Default value is 256. Possible values are:

    • For mistral-7b-instruct-v1, integer values between 1 and 8000.

    • For llama2-7b-v1, integer values between1 and 4096.

    • For llama3-8b-instruct-v1, integer values between 1 and 4096.

  • top_k: specifies the number of top most likely tokens to consider for text generation at each step. Default value is 40, which means that top 40 most likely tokens are considered for text generation at each step. Possible values are integer values between 0 and 32000.

  • top_p: specifies a number, p, and ensures that only the most likely tokens with the sum of probabilities p are considered for generation at each step. A higher value of p introduces more randomness into the output. Default value is 0.95. Possible values are float values between 0 and 1.

    • To disable this method, set to 1.0 or 0.

    • To eliminate tokens with low likelihood, assign p a lower value. For example, if set to 0.1, tokens within top 10% probability are included.

    • To include tokens with low likelihood, assign p a higher value. For example, if set to 0.9, tokens within top 90% probability are included.

    If you are also specifying the top_k parameter, the LLM considers only the top tokens whose probabilities add up to p percent. It ignores the rest of the k tokens.

  • repeat_penalty: assigns a penalty when a token appears repeatedly. High penalties encourage less repeated tokens and produce more random outputs. Default value is 1.1. Possible values are float values between 0 and 2.

  • frequency_penalty: assigns a penalty when a token appears frequently. High penalties encourage less repeated tokens and produce more random outputs. Default value is 0. Possible values are float values between 0 and 1.

  • stop_sequences: specifies a list of characters such as a word, a phrase, a newline, or a period that tells the LLM when to end the generated output. If you have more than one stop sequence, then the LLM stops when it reaches any of those sequences. Default value is NULL.

  • batch_size: specifies the batch size for the routine. This parameter is supported for internal tables only. Default value is 1000. Possible values are integer values between 1 and 1000.

Syntax Examples

• Generating English text-based content in a batch using the mistral-7b-instruct-v1 model:

  call sys.ML_GENERATE_TABLE("demo_db.input_table.Input", "demo_db.output_table.Output", JSON_OBJECT("task", "generation", "model_id", "mistral-7b-instruct-v1", "language", "en"));

  In this example, the routine takes input queries stored in the demo_db.input_table.Input column, and creates a new column demo_db.output_table.Output where it stores the generated text-based responses.

• Summarizing English text in a batch using the mistral-7b-instruct-v1 model:

  call sys.ML_GENERATE_TABLE("demo_db.input_table.Input", "demo_db.output_table.Output", JSON_OBJECT("task", "summarization", "model_id", "mistral-7b-instruct-v1", "language", "en"));

  In this example, the routine takes input queries stored in the demo_db.input_table.Input column, and creates a new column demo_db.output_table.Output where it stores the generated text summaries.