Browse Source

Making old code obsolete - SemanticKernel: Correcting working with PromptExecutionSettings

pull/689/head
Zoli Somogyi 1 year ago
parent
commit
54c01d4c2c
2 changed files with 219 additions and 0 deletions
  1. +114
    -0
      LLama.SemanticKernel/ChatCompletion/ChatRequestSettings.cs
  2. +105
    -0
      LLama.SemanticKernel/ChatCompletion/ChatRequestSettingsConverter.cs

+ 114
- 0
LLama.SemanticKernel/ChatCompletion/ChatRequestSettings.cs View File

@@ -0,0 +1,114 @@
using Microsoft.SemanticKernel;
using System.Text.Json;
using System.Text.Json.Serialization;

namespace LLamaSharp.SemanticKernel.ChatCompletion;

[Obsolete("Use LLamaSharpPromptExecutionSettings instead")]
public class ChatRequestSettings : PromptExecutionSettings
{
/// <summary>
/// Temperature controls the randomness of the completion.
/// The higher the temperature, the more random the completion.
/// </summary>
[JsonPropertyName("temperature")]
public double Temperature { get; set; } = 0;

/// <summary>
/// TopP controls the diversity of the completion.
/// The higher the TopP, the more diverse the completion.
/// </summary>
[JsonPropertyName("top_p")]
public double TopP { get; set; } = 0;

/// <summary>
/// Number between -2.0 and 2.0. Positive values penalize new tokens
/// based on whether they appear in the text so far, increasing the
/// model's likelihood to talk about new topics.
/// </summary>
[JsonPropertyName("presence_penalty")]
public double PresencePenalty { get; set; } = 0;

/// <summary>
/// Number between -2.0 and 2.0. Positive values penalize new tokens
/// based on their existing frequency in the text so far, decreasing
/// the model's likelihood to repeat the same line verbatim.
/// </summary>
[JsonPropertyName("frequency_penalty")]
public double FrequencyPenalty { get; set; } = 0;

/// <summary>
/// Sequences where the completion will stop generating further tokens.
/// </summary>
[JsonPropertyName("stop_sequences")]
public IList<string> StopSequences { get; set; } = Array.Empty<string>();

/// <summary>
/// How many completions to generate for each prompt. Default is 1.
/// Note: Because this parameter generates many completions, it can quickly consume your token quota.
/// Use carefully and ensure that you have reasonable settings for max_tokens and stop.
/// </summary>
[JsonPropertyName("results_per_prompt")]
public int ResultsPerPrompt { get; set; } = 1;

/// <summary>
/// The maximum number of tokens to generate in the completion.
/// </summary>
[JsonPropertyName("max_tokens")]
public int? MaxTokens { get; set; }

/// <summary>
/// Modify the likelihood of specified tokens appearing in the completion.
/// </summary>
[JsonPropertyName("token_selection_biases")]
public IDictionary<int, int> TokenSelectionBiases { get; set; } = new Dictionary<int, int>();

/// <summary>
/// Create a new settings object with the values from another settings object.
/// </summary>
/// <param name="requestSettings">Template configuration</param>
/// <param name="defaultMaxTokens">Default max tokens</param>
/// <returns>An instance of OpenAIRequestSettings</returns>
public static ChatRequestSettings FromRequestSettings(PromptExecutionSettings? requestSettings, int? defaultMaxTokens = null)
{
if (requestSettings is null)
{
return new ChatRequestSettings()
{
MaxTokens = defaultMaxTokens
};
}

if (requestSettings is ChatRequestSettings requestSettingsChatRequestSettings)
{
return requestSettingsChatRequestSettings;
}

var json = JsonSerializer.Serialize(requestSettings);
var chatRequestSettings = JsonSerializer.Deserialize<ChatRequestSettings>(json, s_options);

if (chatRequestSettings is not null)
{
return chatRequestSettings;
}

throw new ArgumentException($"Invalid request settings, cannot convert to {nameof(ChatRequestSettings)}", nameof(requestSettings));
}

private static readonly JsonSerializerOptions s_options = CreateOptions();

private static JsonSerializerOptions CreateOptions()
{
JsonSerializerOptions options = new()
{
WriteIndented = true,
MaxDepth = 20,
AllowTrailingCommas = true,
PropertyNameCaseInsensitive = true,
ReadCommentHandling = JsonCommentHandling.Skip,
Converters = { new ChatRequestSettingsConverter() }
};

return options;
}
}

+ 105
- 0
LLama.SemanticKernel/ChatCompletion/ChatRequestSettingsConverter.cs View File

@@ -0,0 +1,105 @@
using System;
using System.Collections.Generic;
using System.Text.Json;
using System.Text.Json.Serialization;

namespace LLamaSharp.SemanticKernel.ChatCompletion;

/// <summary>
/// JSON converter for <see cref="OpenAIRequestSettings"/>
/// </summary>
[Obsolete("Use LLamaSharpPromptExecutionSettingsConverter instead")]
public class ChatRequestSettingsConverter : JsonConverter<ChatRequestSettings>
{
/// <inheritdoc/>
public override ChatRequestSettings? Read(ref Utf8JsonReader reader, Type typeToConvert, JsonSerializerOptions options)
{
var requestSettings = new ChatRequestSettings();

while (reader.Read() && reader.TokenType != JsonTokenType.EndObject)
{
if (reader.TokenType == JsonTokenType.PropertyName)
{
string? propertyName = reader.GetString();

if (propertyName is not null)
{
// normalise property name to uppercase
propertyName = propertyName.ToUpperInvariant();
}

reader.Read();

switch (propertyName)
{
case "MODELID":
case "MODEL_ID":
requestSettings.ModelId = reader.GetString();
break;
case "TEMPERATURE":
requestSettings.Temperature = reader.GetDouble();
break;
case "TOPP":
case "TOP_P":
requestSettings.TopP = reader.GetDouble();
break;
case "FREQUENCYPENALTY":
case "FREQUENCY_PENALTY":
requestSettings.FrequencyPenalty = reader.GetDouble();
break;
case "PRESENCEPENALTY":
case "PRESENCE_PENALTY":
requestSettings.PresencePenalty = reader.GetDouble();
break;
case "MAXTOKENS":
case "MAX_TOKENS":
requestSettings.MaxTokens = reader.GetInt32();
break;
case "STOPSEQUENCES":
case "STOP_SEQUENCES":
requestSettings.StopSequences = JsonSerializer.Deserialize<IList<string>>(ref reader, options) ?? Array.Empty<string>();
break;
case "RESULTSPERPROMPT":
case "RESULTS_PER_PROMPT":
requestSettings.ResultsPerPrompt = reader.GetInt32();
break;
case "TOKENSELECTIONBIASES":
case "TOKEN_SELECTION_BIASES":
requestSettings.TokenSelectionBiases = JsonSerializer.Deserialize<IDictionary<int, int>>(ref reader, options) ?? new Dictionary<int, int>();
break;
default:
reader.Skip();
break;
}
}
}

return requestSettings;
}

/// <inheritdoc/>
public override void Write(Utf8JsonWriter writer, ChatRequestSettings value, JsonSerializerOptions options)
{
writer.WriteStartObject();

writer.WriteNumber("temperature", value.Temperature);
writer.WriteNumber("top_p", value.TopP);
writer.WriteNumber("frequency_penalty", value.FrequencyPenalty);
writer.WriteNumber("presence_penalty", value.PresencePenalty);
if (value.MaxTokens is null)
{
writer.WriteNull("max_tokens");
}
else
{
writer.WriteNumber("max_tokens", (decimal)value.MaxTokens);
}
writer.WritePropertyName("stop_sequences");
JsonSerializer.Serialize(writer, value.StopSequences, options);
writer.WriteNumber("results_per_prompt", value.ResultsPerPrompt);
writer.WritePropertyName("token_selection_biases");
JsonSerializer.Serialize(writer, value.TokenSelectionBiases, options);

writer.WriteEndObject();
}
}

Loading…
Cancel
Save