Skip to content

Benchmark

BABILong qa1 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

85%
Mean
92%
Median
15.2%
Std dev
59%
10th percentile
100%
90th percentile
57%
Min
57%100%
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-tuneqa1, 1k· BABILong v0 data docs99.3%
~ Mamba (130M) fine-tuneqa1, 1k· BABILong v0 data docs100%
~ RMT (137M) fine-tuneqa1, 1k· BABILong v0 data docs100%
ChatGLM3 6Bchatglm3-6b-128k, qa1, 1k, 128k· BABILong v0 data docs77%
GPT-4-0125-previewqa1, 1k· BABILong v0 data docs100%
Llama 2 7bLLaMA-2-7B-32K, qa1, 1k, 32k· BABILong v0 data docs57%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa1, 1k· BABILong v0 data docs93%
Llama3-ChatQA-1.5-8B + RAGqa1, 1k· BABILong v0 data docs62%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa1, 1k· BABILong v0 data docs86%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa1, 1k· BABILong v0 data docs99%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa1, 1k· BABILong v0 data docs92%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa1, 1k· BABILong v0 data docs93%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa1, 1k· BABILong v0 data docs84%
Yi-34B01-ai/Yi-34B-200k, qa1, 1k, 200k· BABILong v0 data docs73%
Yi-34B01-ai/Yi-34B, qa1, 1k· BABILong v0 data docs59%

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