Skip to content

Benchmark

BABILong qa3 1k

Long context benchmarks measure how well models use information in long inputs. Scores are percentages. Higher is better.

Models and systems measured
14
Scores recorded
15
Sources
1
IQ weight
No weight published

Score distribution

49.2%
Mean
38%
Median
24.1%
Std dev
28%
10th percentile
94%
90th percentile
18%
Min
18%97%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
~ ARMT (137M) fine-tuneqa3, 1k· BABILong v0 data docs85.2%
~ Mamba (130M) fine-tuneqa3, 1k· BABILong v0 data docs97%
~ RMT (137M) fine-tuneqa3, 1k· BABILong v0 data docs94%
ChatGLM3 6Bchatglm3-6b-128k, qa3, 1k, 128k· BABILong v0 data docs37%
GPT-4-0125-previewqa3, 1k· BABILong v0 data docs63%
Llama 2 7bLLaMA-2-7B-32K, qa3, 1k, 32k· BABILong v0 data docs38%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa3, 1k· BABILong v0 data docs28%
Llama3-ChatQA-1.5-8B + RAGqa3, 1k· BABILong v0 data docs18%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa3, 1k· BABILong v0 data docs34%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa3, 1k· BABILong v0 data docs56%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa3, 1k· BABILong v0 data docs32%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa3, 1k· BABILong v0 data docs51%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa3, 1k· BABILong v0 data docs41%
Yi-34B01-ai/Yi-34B-200k, qa3, 1k, 200k· BABILong v0 data docs30%
Yi-34B01-ai/Yi-34B, qa3, 1k· BABILong v0 data docs34%

Data credits: RMT-team, Kuratov et al. Official v0 numeric CSV; results licence not separately stated.