Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Less Sampler Allocations #735

Merged
merged 5 commits into from
May 14, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 10 additions & 9 deletions LLama.Examples/Examples/BatchedExecutorGuidance.cs
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
using LLama.Batched;
using LLama.Batched;
using LLama.Common;
using LLama.Native;
using LLama.Sampling;
Expand Down Expand Up @@ -105,18 +105,19 @@ public override ISamplingPipeline Clone()

protected override void ProcessLogits(SafeLLamaContextHandle ctx, Span<float> logits, ReadOnlySpan<LLamaToken> lastTokens)
{
if (guidance == null)
return;

// Get the logits generated by the guidance sequences
var guidanceLogits = guidance.Sample();

// Use those logits to guide this sequence
NativeApi.llama_sample_apply_guidance(ctx, logits, guidanceLogits, weight);
}

protected override LLamaToken ProcessTokenDataArray(SafeLLamaContextHandle ctx, LLamaTokenDataArray candidates, ReadOnlySpan<LLamaToken> lastTokens)
{
if (guidance != null)
{
// Get the logits generated by the guidance sequences
var guidanceLogits = guidance.Sample();

// Modify these logits based on the guidance logits
candidates.Guidance(ctx, guidanceLogits, weight);
}

candidates.Temperature(ctx, 0.8f);
candidates.TopK(ctx, 25);

Expand Down
12 changes: 8 additions & 4 deletions LLama/LLamaContext.cs
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,8 @@ public uint BatchThreads
/// Get the maximum batch size for this context
/// </summary>
public uint BatchSize => NativeHandle.BatchSize;

private LLamaTokenData[]? _samplingBuffer;

/// <summary>
/// Create a new LLamaContext for the given LLamaWeights
Expand Down Expand Up @@ -496,7 +498,9 @@ public LLamaTokenDataArray ApplyPenalty(int logits_i, IEnumerable<LLamaToken> la
var nl_logit = logits[(int?)nl_token ?? 0];

// Convert logits into token candidates
var candidates_p = LLamaTokenDataArray.Create(logits);
if (_samplingBuffer == null || _samplingBuffer.Length < logits.Length)
_samplingBuffer = new LLamaTokenData[logits.Length];
var candidates_p = LLamaTokenDataArray.Create(logits, _samplingBuffer);

// Extract most recently returned tokens
var last_n_repeat = Math.Min((int)ContextSize, repeatLastTokensCount);
Expand All @@ -508,14 +512,14 @@ public LLamaTokenDataArray ApplyPenalty(int logits_i, IEnumerable<LLamaToken> la
// Restore newline token logit value if necessary
if (!penalizeNL && nl_token.HasValue)
{
var candidatesSpan = candidates_p.data.Span;
for (var i = 0; i < candidates_p.data.Length; i++)
var candidatesSpan = candidates_p.Data.Span;
for (var i = 0; i < candidates_p.Data.Length; i++)
{
ref var item = ref candidatesSpan[i];
if (item.id == nl_token)
item.logit = nl_logit;
}
candidates_p.sorted = false;
candidates_p.Sorted = false;
}

return candidates_p;
Expand Down
Loading
Loading