Llama 3 8B 4bit UltraChat Ita: benchmark results
FairMind's Italian-language fine-tune of Meta's Llama 3 8B, trained in 4-bit with Unsloth on Italian-translated chat data. The published download contains BF16 weights; 4bit in the model name refers to training. Provider: walid-iguider. Released 2024-04-18. Access: Open.
Aggregate Rating 1383 ± 22, rank #1280 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EVALITA - word-in-context | 66.22 | CPS | 70.2 |
| Open Italian LLM - MMLU-Pro (IT) | 26.01 | Accuracy (%) | 67.9 |
| Open LLM Leaderboard v1 - MMLU | 63.85 | Accuracy (%) (5-shot) | 63.7 |
| Open LLM Leaderboard v1 - WinoGrande | 77.43 | Accuracy (%) (5-shot) | 55.8 |
| Open LLM Leaderboard v1 - GSM8K | 39.35 | Accuracy (%) (5-shot) | 54.3 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 48.88 | MC2 (%) (0-shot) | 43 |
| Open LLM Leaderboard v1 - ARC Challenge | 59.04 | Normalized accuracy (%) (25-shot) | 41.9 |
| Open LLM Leaderboard v1 - HellaSwag | 80.67 | Normalized accuracy (%) (10-shot) | 41.8 |
| Pinocchio Italian - Cultura | 51.6 | Accuracy (%) | 40.9 |
| EVALITA - faq | 52.88 | CPS | 40.4 |
| EVALITA - sentiment-analysis | 69.44 | CPS | 38.3 |
| Pinocchio Italian - Diritto | 44 | Accuracy (%) | 36.4 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-4bit-ultrachat-ita · How It Works · Data refreshed daily, snapshot 2026-10-20.