LLamaSharp/LLama.Unittest/BeamTests.cs

73 lines
2.2 KiB
C#

using System.Text;
using LLama.Common;
using LLama.Native;
using Xunit.Abstractions;
namespace LLama.Unittest;
public sealed class BeamTests
: IDisposable
{
private readonly ITestOutputHelper _testOutputHelper;
private readonly ModelParams _params;
private readonly LLamaWeights _model;
public BeamTests(ITestOutputHelper testOutputHelper)
{
_testOutputHelper = testOutputHelper;
_params = new ModelParams(Constants.GenerativeModelPath)
{
ContextSize = 2048,
GpuLayerCount = Constants.CIGpuLayerCount,
};
_model = LLamaWeights.LoadFromFile(_params);
}
public void Dispose()
{
_model.Dispose();
}
[Fact]
public void BasicBeam()
{
const int num_beams = 2;
const int n_predict = 3;
const string prompt = "The cat sat on";
var context = _model.CreateContext(_params);
var initial_tokens = context.Tokenize(prompt);
var batch = new LLamaBatch();
batch.AddRange(initial_tokens, 0, LLamaSeqId.Zero, true);
context.Decode(batch);
var decoder = new StreamingTokenDecoder(context);
NativeApi.llama_beam_search(context.NativeHandle, (data, state) =>
{
// Show the current state of every beam.
for (var i = 0; i < state.Beams.Length; i++)
{
ref var view = ref state.Beams[i];
var decoder = new StreamingTokenDecoder(context);
decoder.AddRange(view.Tokens);
var tokens = decoder.Read();
_testOutputHelper.WriteLine($"B{i} ({view.CumulativeProbability}) => '{tokens}'");
}
// Once all beams agree on some tokens read them and append them to the output decoder
if (state.CommonPrefixLength > 0)
{
var view = state.Beams[0];
decoder.AddRange(view.Tokens.Slice(0, (int)state.CommonPrefixLength));
}
}, IntPtr.Zero, num_beams, initial_tokens.Length, n_predict, Math.Max(1, Environment.ProcessorCount / 2));
_testOutputHelper.WriteLine($"Final: {prompt}{decoder.Read()}");
}
}