Method Drift›Long-context / context-window extension
MInference
Long-context / context-window extension
superseded — cited as a baseline and beaten by newer methods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”
— Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs
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 for Efficient Long Context Modeling
Latent-Condensed Transformer beats MInference · Avg. [64K context on H200 GPUs]
29.09 vs 19.71
- Latent-Condensed Transformer for Efficient Long Context Modeling
Latent-Condensed Transformer beats MInference · Avg. [128K context on H200 GPUs]
58.80 vs 37.60
- Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs
DHSA beats MInference · Avg. [Llama-3.1-8B-Instruct (4-bit)]
31.8 vs 28.4
- Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs
DHSA beats MInference · 32K [32K tokens]
76.2 vs 74.7
- Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs
DHSA beats MInference · 48K [48K tokens]
71.5 vs 64.2
- Curse of High Dimensionality Issue in Transformer for Long-context Modeling
DGA-LLM (Ours) beats MInference · Inter-Token Latency [LongBench-E benchmark]
28.80 vs 94.34
- Curse of High Dimensionality Issue in Transformer for Long-context Modeling
DGA-LLM (Ours) beats MInference · Avg. EM Score [EM evaluation at 4K-32K lengths]
27.7 vs 23.7
- Training-free Context-adaptive Attention for Efficient Long Context Modeling
SexyName beats MInference · MMLU [LLaMA3.1-8B-Instruct]
69.21 vs 69.14
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.