Skip to content

Benchmark

BABILong qa3 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

45.5%
Mean
31%
Median
25.1%
Std dev
24%
10th percentile
90.5%
90th percentile
17%
Min
17%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, 2k· BABILong v0 data docs90.5%
~ Mamba (130M) fine-tuneqa3, 2k· BABILong v0 data docs97%
~ RMT (137M) fine-tuneqa3, 2k· BABILong v0 data docs88%
ChatGLM3 6Bchatglm3-6b-128k, qa3, 2k, 128k· BABILong v0 data docs31%
GPT-4-0125-previewqa3, 2k· BABILong v0 data docs57%
Llama 2 7bLLaMA-2-7B-32K, qa3, 2k, 32k· BABILong v0 data docs34%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa3, 2k· BABILong v0 data docs30%
Llama3-ChatQA-1.5-8B + RAGqa3, 2k· BABILong v0 data docs17%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa3, 2k· BABILong v0 data docs31%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa3, 2k· BABILong v0 data docs49%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa3, 2k· BABILong v0 data docs31%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa3, 2k· BABILong v0 data docs45%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa3, 2k· BABILong v0 data docs31%
Yi-34B01-ai/Yi-34B-200k, qa3, 2k, 200k· BABILong v0 data docs27%
Yi-34B01-ai/Yi-34B, qa3, 2k· BABILong v0 data docs24%

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