Skip to content

Benchmark

BABILong qa4 8k

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

62.2%
Mean
60%
Median
17.7%
Std dev
46%
10th percentile
100%
90th percentile
35%
Min
35%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, 8k· BABILong v0 data docs100%
~ Mamba (130M) fine-tuneqa4, 8k· BABILong v0 data docs100%
~ RMT (137M) fine-tuneqa4, 8k· BABILong v0 data docs80%
ChatGLM3 6Bchatglm3-6b-128k, qa4, 8k, 128k· BABILong v0 data docs35%
GPT-4-0125-previewqa4, 8k· BABILong v0 data docs52%
Llama 2 7bLLaMA-2-7B-32K, qa4, 8k, 32k· BABILong v0 data docs46%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa4, 8k· BABILong v0 data docs52%
Llama3-ChatQA-1.5-8B + RAGqa4, 8k· BABILong v0 data docs57%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa4, 8k· BABILong v0 data docs60%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa4, 8k· BABILong v0 data docs62%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa4, 8k· BABILong v0 data docs63%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa4, 8k· BABILong v0 data docs64%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa4, 8k· BABILong v0 data docs49%
Yi-34B01-ai/Yi-34B-200k, qa4, 8k, 200k· BABILong v0 data docs61%
Yi-34B01-ai/Yi-34B, qa4, 8k· BABILong v0 data docs52%

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