Skip to content

Benchmark

BABILong qa5 2k

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

79.9%
Mean
78%
Median
12.4%
Std dev
64%
10th percentile
99%
90th percentile
59%
Min
59%99.1%
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-tuneqa5, 2k· BABILong v0 data docs99.1%
~ Mamba (130M) fine-tuneqa5, 2k· BABILong v0 data docs98%
~ RMT (137M) fine-tuneqa5, 2k· BABILong v0 data docs99%
ChatGLM3 6Bchatglm3-6b-128k, qa5, 2k, 128k· BABILong v0 data docs64%
GPT-4-0125-previewqa5, 2k· BABILong v0 data docs92%
Llama 2 7bLLaMA-2-7B-32K, qa5, 2k, 32k· BABILong v0 data docs70%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa5, 2k· BABILong v0 data docs73%
Llama3-ChatQA-1.5-8B + RAGqa5, 2k· BABILong v0 data docs78%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa5, 2k· BABILong v0 data docs66%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa5, 2k· BABILong v0 data docs84%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa5, 2k· BABILong v0 data docs80%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa5, 2k· BABILong v0 data docs84%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa5, 2k· BABILong v0 data docs76%
Yi-34B01-ai/Yi-34B-200k, qa5, 2k, 200k· BABILong v0 data docs77%
Yi-34B01-ai/Yi-34B, qa5, 2k· BABILong v0 data docs59%

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