Atlas

Models

← All models

Qwen2.5 3B

Alibaba · 2024-09-19 · 3.1B parameters

Qwen2.5-3B is a base (pretrained) checkpoint of Alibaba's Qwen2.5 series, released on September 19, 2024 under the Qwen Research license. It is a dense decoder-only Transformer with 3,085,938,688 parameters, grouped-query attention, tied embeddings and a 32,768-token context length.

Benchmark scores

BenchmarkScore
ARC-Challenge56.5
BBH (BIG-Bench Hard)56.3
Capability100.3
GPQA - release-reported, subset unspecified26.3
GSM8K79.1
HellaSwag (Unspecified Scoring Protocol)74.6
HumanEval42.1
HumanEval+36.0
MATH42.6
MBPP57.1
MBPP+49.4
MMLU65.6
MMLU Pro34.6
MMLU STEM62.5
MMLU-Redux63.7
MultiPL-E (HumanEval, 8-language average)41.2
TheoremQA27.4
TruthfulQA (Unspecified Metric)48.9
WinoGrande71.1
Loading Atlas data…