Skip to content

Benchmark

BABILong qa2 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

45.4%
Mean
35%
Median
29.5%
Std dev
14%
10th percentile
98%
90th percentile
11%
Min
11%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, 8k· BABILong v0 data docs100%
~ Mamba (130M) fine-tuneqa2, 8k· BABILong v0 data docs98%
~ RMT (137M) fine-tuneqa2, 8k· BABILong v0 data docs97%
ChatGLM3 6Bchatglm3-6b-128k, qa2, 8k, 128k· BABILong v0 data docs38%
GPT-4-0125-previewqa2, 8k· BABILong v0 data docs65%
Llama 2 7bLLaMA-2-7B-32K, qa2, 8k, 32k· BABILong v0 data docs27%
Llama 3 8B InstructMeta-Llama-3-8B-Instruct, qa2, 8k· BABILong v0 data docs20%
Llama3-ChatQA-1.5-8B + RAGqa2, 8k· BABILong v0 data docs14%
Mistral 7B Instruct v0.2Mistral-7b-Instruct-v0.2, qa2, 8k· BABILong v0 data docs11%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, qa2, 8k· BABILong v0 data docs48%
Mixtral 8x7B Instruct v0.1Mixtral-8x7B-Instruct-v0.1, qa2, 8k· BABILong v0 data docs27%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, qa2, 8k· BABILong v0 data docs44%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, qa2, 8k· BABILong v0 data docs34%
Yi-34B01-ai/Yi-34B-200k, qa2, 8k, 200k· BABILong v0 data docs35%
Yi-34B01-ai/Yi-34B, qa2, 8k· BABILong v0 data docs23%

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