Skip to content

Benchmark

BABILong qa5 32k

Long context benchmarks measure how well models use information in long inputs. Scores are percentages. Higher is better.

Models and systems measured
13
Scores recorded
14
Sources
1
IQ weight
No weight published

Score distribution

66.4%
Mean
76%
Median
27.9%
Std dev
7%
10th percentile
98%
90th percentile
4%
Min
4%98%
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, 32k· BABILong v0 data docs98%
~ Mamba (130M) fine-tuneqa5, 32k· BABILong v0 data docs98%
~ RMT (137M) fine-tuneqa5, 32k· BABILong v0 data docs90%
ChatGLM3 6Bchatglm3-6b-128k, qa5, 32k, 128k· BABILong v0 data docs58%
GPT-4-0125-previewqa5, 32k· BABILong v0 data docs82%
Llama 2 7bLLaMA-2-7B-32K, qa5, 32k· BABILong v0 data docs7%
Llama3-ChatQA-1.5-8B + RAGqa5, 32k· BABILong v0 data docs78%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa5, 32k· BABILong v0 data docs67%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa5, 32k· BABILong v0 data docs63%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa5, 32k· BABILong v0 data docs56%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa5, 32k· BABILong v0 data docs81%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa5, 32k· BABILong v0 data docs71%
Yi-34B01-ai/Yi-34B-200k, qa5, 32k, 200k· BABILong v0 data docs76%
Yi-34B01-ai/Yi-34B, qa5, 32k· BABILong v0 data docs4%

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