Papers · Proving inference · Lu et al.
Lu et al.
Efficient Zero-Knowledge Proofs for LLM Inference (VOLE-based)
Reported benchmarks
| model | params | proving time s | proof size mb | verification time s |
|---|---|---|---|---|
| GPT-2 | 117M | 287.1 | 4500 | 63 |
Notes
CONFLICT: survey Table VI reports 287.1s / 63s / 4.5GB for GPT-2; zkGPT's Table 3 measures 112.3s / 31.4s / 2.24GB (32 threads, incl. communication). Designated-verifier and huge proofs -- the tradeoff VOLE buys for prover speed.
Our reading
Citation neighbourhood
Builds on
- none recorded
Cited by, in this corpus
- none recorded
Edges are a proxy: paper A's text mentions B anywhere (body or bibliography). See the full graph.
Discussed in
Two dialectsThe prior surveys, and where we disagree with themSystem by systemThe landscapeProof systemsWhat is actually being proven
Other recorded fields
claim_kind: pass