Skip to content

Benchmark

BABILong qa4 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

63.3%
Mean
60%
Median
17.5%
Std dev
43%
10th percentile
100%
90th percentile
38%
Min
38%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, 4k· BABILong v0 data docs100%
~ Mamba (130M) fine-tuneqa4, 4k· BABILong v0 data docs100%
~ RMT (137M) fine-tuneqa4, 4k· BABILong v0 data docs83%
ChatGLM3 6Bchatglm3-6b-128k, qa4, 4k, 128k· BABILong v0 data docs38%
GPT-4-0125-previewqa4, 4k· BABILong v0 data docs60%
Llama 2 7bLLaMA-2-7B-32K, qa4, 4k, 32k· BABILong v0 data docs51%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa4, 4k· BABILong v0 data docs43%
Llama3-ChatQA-1.5-8B + RAGqa4, 4k· BABILong v0 data docs59%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa4, 4k· BABILong v0 data docs60%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa4, 4k· BABILong v0 data docs62%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa4, 4k· BABILong v0 data docs61%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa4, 4k· BABILong v0 data docs64%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa4, 4k· BABILong v0 data docs50%
Yi-34B01-ai/Yi-34B-200k, qa4, 4k, 200k· BABILong v0 data docs63%
Yi-34B01-ai/Yi-34B, qa4, 4k· BABILong v0 data docs55%

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