Skip to content

Benchmark

BABILong qa2 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

50.6%
Mean
39%
Median
27%
Std dev
22%
10th percentile
98%
90th percentile
19%
Min
19%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-tuneqa2, 4k· BABILong v0 data docs100%
~ Mamba (130M) fine-tuneqa2, 4k· BABILong v0 data docs98%
~ RMT (137M) fine-tuneqa2, 4k· BABILong v0 data docs98%
ChatGLM3 6Bchatglm3-6b-128k, qa2, 4k, 128k· BABILong v0 data docs39%
GPT-4-0125-previewqa2, 4k· BABILong v0 data docs68%
Llama 2 7bLLaMA-2-7B-32K, qa2, 4k, 32k· BABILong v0 data docs26%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa2, 4k· BABILong v0 data docs39%
Llama3-ChatQA-1.5-8B + RAGqa2, 4k· BABILong v0 data docs19%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa2, 4k· BABILong v0 data docs22%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa2, 4k· BABILong v0 data docs56%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa2, 4k· BABILong v0 data docs35%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa2, 4k· BABILong v0 data docs51%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa2, 4k· BABILong v0 data docs36%
Yi-34B01-ai/Yi-34B-200k, qa2, 4k, 200k· BABILong v0 data docs42%
Yi-34B01-ai/Yi-34B, qa2, 4k· BABILong v0 data docs30%

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