blossom-v5-4B: benchmark results
A conversational instruction fine-tune released by Azure99 based on Qwen1.5-4B, using Azure99/blossom-chat-v3, Azure99/blossom-math-v4, Azure99/blossom-wizard-v3, Azure99/blossom-orca-v3. Training uses two stages; dataset example counts and epochs are not token or compute disclosures. The card says its training data was distilled from gpt-4-0125-preview. Provider: Azure99. Access: Open.
Aggregate Rating 1329 ± 38, rank #8174 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 48.37 | Accuracy (%) (5-shot) | 61.6 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 47.51 | MC2 (%) (0-shot) | 38 |
| Open LLM Leaderboard v1 - MMLU | 55.04 | Accuracy (%) (5-shot) | 35.3 |
| Open LLM Leaderboard v1 - HellaSwag | 71.87 | Normalized accuracy (%) (10-shot) | 22.1 |
| Open LLM Leaderboard v1 - ARC Challenge | 46.76 | Normalized accuracy (%) (25-shot) | 21.8 |
| Open LLM Leaderboard v1 - WinoGrande | 67.4 | Accuracy (%) (5-shot) | 21.4 |
Interactive version: theaggregate.ai/model?slug=blossom-v5-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.