mm4-3B: benchmark results
mm4-3b is a liminerity release configured as a 40-layer Llama causal language model. Its card describes training and merging but does not supply a reproducible exact recipe; no total training-token or parameter count is inferred from its name. Provider: liminerity. Access: Open.
Aggregate Rating 1305 ± 38, rank #8649 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 43.82 | Accuracy (%) (5-shot) | 58 |
| Open LLM Leaderboard v1 - MMLU | 50.9 | Accuracy (%) (5-shot) | 30.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 43.2 | MC2 (%) (0-shot) | 22.9 |
| Open LLM Leaderboard v1 - HellaSwag | 70.41 | Normalized accuracy (%) (10-shot) | 20.7 |
| Open LLM Leaderboard v1 - ARC Challenge | 44.8 | Normalized accuracy (%) (25-shot) | 20.4 |
| Open LLM Leaderboard v1 - WinoGrande | 66.22 | Accuracy (%) (5-shot) | 19.9 |
Interactive version: theaggregate.ai/model?slug=mm4-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.