|
| 1 | +using LLama.Common; |
| 2 | +using LLamaSharp.KernelMemory; |
| 3 | +using Microsoft.KernelMemory.AI; |
| 4 | +using Xunit.Abstractions; |
| 5 | + |
| 6 | +namespace LLama.Unittest.KernelMemory |
| 7 | +{ |
| 8 | + |
| 9 | + public abstract class ITextTokenizerTests |
| 10 | + { |
| 11 | + private readonly ITestOutputHelper _testOutputHelper; |
| 12 | + |
| 13 | +#pragma warning disable KMEXP00 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed. |
| 14 | + protected ITextTokenizer? _generator; |
| 15 | +#pragma warning restore KMEXP00 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed. |
| 16 | + |
| 17 | + protected InferenceParams _infParams; |
| 18 | + protected LLamaSharpConfig _lsConfig; |
| 19 | + |
| 20 | + public ITextTokenizerTests(ITestOutputHelper testOutputHelper) |
| 21 | + { |
| 22 | + _testOutputHelper = testOutputHelper; |
| 23 | + |
| 24 | + _infParams = new() { AntiPrompts = ["\n\n"] }; |
| 25 | + _lsConfig = new(Constants.GenerativeModelPath) { DefaultInferenceParams = _infParams }; |
| 26 | + |
| 27 | + testOutputHelper.WriteLine($"Using model {Path.GetFileName(_lsConfig.ModelPath)}"); |
| 28 | + } |
| 29 | + |
| 30 | + |
| 31 | + [Theory] |
| 32 | + [InlineData("The quick brown fox jumps over the lazy dog")] |
| 33 | + [InlineData("Well, here're some special characters!!!")] |
| 34 | + [InlineData("...___---")] |
| 35 | + [InlineData("15 + 6 = 21 && 68 * 75 = 5100")] |
| 36 | + [InlineData(" \n \r\n \t ")] |
| 37 | + public void GetTokens_ShouldReturnListOfTokensForInputString(string? text) |
| 38 | + { |
| 39 | + var tokens = _generator!.GetTokens(text); |
| 40 | + var tokensCount = _generator.CountTokens(text); |
| 41 | + |
| 42 | + var expected = " " + text; // the placement of the space corresponding to BOS will vary by model tokenizer |
| 43 | + var actual = string.Join("", tokens); |
| 44 | + |
| 45 | + _testOutputHelper.WriteLine($"Tokens for '{text}':"); |
| 46 | + _testOutputHelper.WriteLine(string.Join("", tokens.Select(x => $"({x})"))); |
| 47 | + |
| 48 | + Assert.Equal(expected, actual); |
| 49 | + Assert.Equal(tokensCount, tokens.Count); |
| 50 | + } |
| 51 | + |
| 52 | + /* This is exactly the same test as the non-unicode cases. However, there are reasons why this |
| 53 | + * should be made a special case and may deviate in the future: |
| 54 | + * |
| 55 | + * As of now there appears to be no final word as to how characters that consist of more than one |
| 56 | + * numeric token should correspond to textual tokens, and results vary according to different |
| 57 | + * models' tokenizers. For example, given a character 'Z' that corresponds to the numeric tokens {1,2,3} |
| 58 | + * some (llama-2) will pad the length of the total number of tokens by returning spaces as tokens |
| 59 | + * (i.e. ' ', ' ', 'Z') while others (GPT4Tokenizer) will pad with the character itself (i.e. 'Z','Z','Z'). |
| 60 | + * |
| 61 | + * This is very evident when tokenizing ideograms and emojis, but can arise with various unicode characters |
| 62 | + * as well. See pull request for more relevant discussion https://github.com/SciSharp/LLamaSharp/pull/862 |
| 63 | + * |
| 64 | + * Currently the method will remain consistent with the output of ITextTokenizer.CountTokens, meaning |
| 65 | + * any redundant tokens will not be omitted as long as they are counted by CountTokens. |
| 66 | + * |
| 67 | + * StreamingTokenDecoder, while sufficiently useful for this task, was not designed with producing |
| 68 | + * output for one numeric token at a time in mind, so ITextTokenizer.GetTokens should not be considered |
| 69 | + * an example of proper use. |
| 70 | + * |
| 71 | + * Note: if this message is removed, also remove references to it in LLamaSharpTextEmbeddingGenerator.GetTokens |
| 72 | + * and LLamaSharpTextGenerator.GetTokens |
| 73 | + */ |
| 74 | + [Theory] |
| 75 | + [InlineData("And a little bit of unicode για να κρατήσουμε τα πράγματα ενδιαφέροντα")] |
| 76 | + [InlineData("猫坐在垫子上 😀🤨🤐😏")] |
| 77 | + public void GetTokens_Unicode_ShouldReturnListOfTokensForInputString(string? text) |
| 78 | + { |
| 79 | + var tokens = _generator!.GetTokens(text); |
| 80 | + var tokensCount = _generator.CountTokens(text); |
| 81 | + |
| 82 | + var expected = " " + text; // the placement of the space corresponding to BOS will vary by model tokenizer |
| 83 | + var actual = string.Join("", tokens); |
| 84 | + |
| 85 | + _testOutputHelper.WriteLine($"Tokens for '{text}':"); |
| 86 | + _testOutputHelper.WriteLine(string.Join("", tokens.Select(x => $"({x})"))); |
| 87 | + |
| 88 | + Assert.Equal(expected, actual); |
| 89 | + Assert.Equal(tokensCount, tokens.Count); |
| 90 | + } |
| 91 | + |
| 92 | + [Fact] |
| 93 | + public void GetToken_ShouldThrowForNull() |
| 94 | + { |
| 95 | + string? text = null; |
| 96 | + |
| 97 | + Assert.Throws<ArgumentNullException>(() => { _generator!.GetTokens(text!); }); |
| 98 | + } |
| 99 | + |
| 100 | + [Fact] |
| 101 | + public void GetToken_EmptyStringYieldsOneEmptyToken() |
| 102 | + { |
| 103 | + var text = ""; |
| 104 | + var expected = ""; |
| 105 | + |
| 106 | + var tokens = _generator!.GetTokens(text); |
| 107 | + var tokensCount = _generator.CountTokens(text); |
| 108 | + var actual = tokens.Single(); |
| 109 | + |
| 110 | + _testOutputHelper.WriteLine($"Tokens for '{text}':"); |
| 111 | + _testOutputHelper.WriteLine(string.Join("", tokens.Select(x => $"({x})"))); |
| 112 | + |
| 113 | + Assert.Equal(expected, actual); |
| 114 | + Assert.Equal(tokensCount, tokens.Count); |
| 115 | + } |
| 116 | + } |
| 117 | +} |
0 commit comments