Skip to content

Benchmark

BABILong qa4 1k

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

66.4%
Mean
60%
Median
16.5%
Std dev
54%
10th percentile
99.4%
90th percentile
48%
Min
48%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-tuneqa4, 1k· BABILong v0 data docs99.4%
~ Mamba (130M) fine-tuneqa4, 1k· BABILong v0 data docs100%
~ RMT (137M) fine-tuneqa4, 1k· BABILong v0 data docs94%
ChatGLM3 6Bchatglm3-6b-128k, qa4, 1k, 128k· BABILong v0 data docs48%
GPT-4-0125-previewqa4, 1k· BABILong v0 data docs70%
Llama 2 7bLLaMA-2-7B-32K, qa4, 1k, 32k· BABILong v0 data docs54%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa4, 1k· BABILong v0 data docs55%
Llama3-ChatQA-1.5-8B + RAGqa4, 1k· BABILong v0 data docs58%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa4, 1k· BABILong v0 data docs58%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa4, 1k· BABILong v0 data docs62%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa4, 1k· BABILong v0 data docs60%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa4, 1k· BABILong v0 data docs61%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa4, 1k· BABILong v0 data docs56%
Yi-34B01-ai/Yi-34B-200k, qa4, 1k, 200k· BABILong v0 data docs65%
Yi-34B01-ai/Yi-34B, qa4, 1k· BABILong v0 data docs56%

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