Long-context / context-window extension
MInference
Superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 4 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites MInference as a baseline.
both rely on manually designed patterns or rules, which limits their ability to capture highly input-dependent attention sparsity
Beaten on benchmarks
Head-to-head results where a newer method reports beating MInference. Values are copied from the source paper's tables — verify against the cited paper.
Latent-Condensed Transformer beats MInference
58.80 vs 37.60
Avg. · [128K context on H200 GPUs]
Latent-Condensed Transformer for Efficient Long Context ModelingDHSA beats MInference
31.8 vs 28.4
Avg. · [Llama-3.1-8B-Instruct (4-bit)]
Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMsSexyName beats MInference
69.21 vs 69.14
MMLU · [LLaMA3.1-8B-Instruct]
Training-free Context-adaptive Attention for Efficient Long Context Modeling
What to use instead
Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.