Method Drift›Speculative decoding
EdgeLLM
EDGE-LLM: Enabling Efficient Large Language Model Adaptation on Edge Devices via Layerwise Unified Compression and Adaptive Layer Tuning and VotingSpeculative decoding · first seen Jun 22, 2024
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 1 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites EdgeLLM as a baseline.
“EdgeLLM xu2024edgellm employs local parallel tree generation but imposes heavy memory and computing burdens on limited hardware.”
— A Pipelined Collaborative Speculative Decoding Framework for Efficient Edge-Cloud LLM Inference
Beaten on benchmarks
Head-to-head results where a newer method reports beating EdgeLLM. Values are copied from the source paper's tables — verify against the cited paper.
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 1, HumanEval]
129 vs 153
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 1, GSM8K]
145 vs 169
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 2, HumanEval]
134 vs 166
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 2, GSM8K]
168 vs 197
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 3, HumanEval]
152 vs 201
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 3, GSM8K]
186 vs 231
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 4, HumanEval]
108 vs 127
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · TPT [Scenario 4, GSM8K]
139 vs 161
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · ECS [HumanEval]
56 vs 75
- PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding
PipeSD beats EdgeLLM · ECS [GSM8K]
84 vs 100
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.