Compare runs

Pick two or more runs to compare. Llama3.2-Instruct on M3 Pro 18 GB is the first; add another below.

No other run of Llama3.2-Instruct or on M3 Pro 18 GB to add yet.

The runs side by side: score, tasks passed, speed, tokens read and written, each category and each task
RunA check marks the best of each row.Llama3.2-InstructM3 Pro 18 GB · llama.cppRemove Llama3.2-Instruct
Score2,170Silver
Tasks passed15 of 24
Speed34.7 tok/s
Tokens5,132
Summarizing3/5
One-sentence summaryPassed33.9 tok/s
Meeting notes in three bulletsPassed33.5 tok/s
Review sentiment and complaintFailed35.5 tok/s
Faithful summary of a studyPassed33.5 tok/s
Action items from an email threadFailed34 tok/s
Extracting2/5
Fields from an email as JSONFailed32.9 tok/s
Every email address, oncePassed35.3 tok/s
A list as CSVFailed34.9 tok/s
The real deadlinePassed39.6 tok/s
Warnings and errors from a logFailed34.7 tok/s
Following instructions5/5
Exactly one wordPassed63.9 tok/s
Three lowercase linesPassed35.1 tok/s
Exact JSON shapePassed35.8 tok/s
Sort a listPassed36.7 tok/s
A name with three rulesPassed40.4 tok/s
Reasoning3/5
Unit pricePassed34.8 tok/s
Day and dateFailed35 tok/s
Who owns each petFailed34.8 tok/s
Arrival timePassed35.1 tok/s
Filling a tankPassed34.7 tok/s
Reading code2/4
What does it printPassed34.8 tok/s
Trace a loopFailed34.6 tok/s
Find the bugPassed35.2 tok/s
Run a query in your headFailed36.7 tok/s