CERBERUS/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-08-04 14:56:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

mlx: add memory trace logging (#16131)

Browse source 
This should help narrow down the root cause of #16030
This commit is contained in:
Daniel Hiltgen 2026-05-13 13:37:31 -07:00 • committed by GitHub
parent 3af1a008e2
commit 6398cd5b78
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

5
x/mlxrunner/pipeline.go
View file

@ -133,6 +133,7 @@ func (r *Runner) TextGenerationPipeline(ctx context.Context, request Request) er
processed += n
position += n
slog.Info("Prompt processing progress", "processed", processed, "total", total)
logutil.TraceContext(ctx, "mlx prompt forward", "processed", processed, "total", total, "tokens", n, "memory", mlx.Memory{})
// Create snapshot if we've reached a pending offset.
if snapOffset := session.nextPendingSnapshot(); snapOffset > 0 {
@ -171,6 +172,7 @@ func (r *Runner) TextGenerationPipeline(ctx context.Context, request Request) er
}
sample = step(mlx.FromValues(tokens[processed:], 1, total-processed))
logutil.TraceContext(ctx, "mlx decode seed", "tokens", total-processed, "memory", mlx.Memory{})
dec := decoder{
tokenizer: r.Tokenizer,
@ -194,6 +196,9 @@ func (r *Runner) TextGenerationPipeline(ctx context.Context, request Request) er
output := int32(sample.Token.Int())
session.outputs = append(session.outputs, output)
if i == 0 {
logutil.TraceContext(ctx, "mlx decode first token", "memory", mlx.Memory{})
}
if r.Tokenizer.IsEOS(output) {
final.DoneReason = 0