Skip to content

Benchmark

BABILong qa5 4k

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

76.7%
Mean
76%
Median
14.4%
Std dev
60%
10th percentile
99%
90th percentile
48%
Min
48%99%
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, 4k· BABILong v0 data docs98.1%
~ Mamba (130M) fine-tuneqa5, 4k· BABILong v0 data docs99%
~ RMT (137M) fine-tuneqa5, 4k· BABILong v0 data docs99%
ChatGLM3 6Bchatglm3-6b-128k, qa5, 4k, 128k· BABILong v0 data docs60%
GPT-4-0125-previewqa5, 4k· BABILong v0 data docs90%
Llama 2 7bLLaMA-2-7B-32K, qa5, 4k, 32k· BABILong v0 data docs67%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa5, 4k· BABILong v0 data docs65%
Llama3-ChatQA-1.5-8B + RAGqa5, 4k· BABILong v0 data docs77%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa5, 4k· BABILong v0 data docs67%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa5, 4k· BABILong v0 data docs79%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa5, 4k· BABILong v0 data docs73%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa5, 4k· BABILong v0 data docs81%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa5, 4k· BABILong v0 data docs72%
Yi-34B01-ai/Yi-34B-200k, qa5, 4k, 200k· BABILong v0 data docs76%
Yi-34B01-ai/Yi-34B, qa5, 4k· BABILong v0 data docs48%

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