File size: 3,586 Bytes
4ca4e4c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 | {
"schema_version": 1,
"title": "Reproduction audit: Hybrid Sequence Models",
"emoji": "\ud83d\udd2c",
"proposed_target_slug": "repro-hybrid-seq-82ejxjzg6r",
"publication_status": "local-only; no Space created or modified",
"paper": {
"arxiv_id": "2603.08859v1",
"openreview_id": "82EJxJzG6r"
},
"updated_at": "2026-07-28T00:00:00+00:00",
"root": {
"slug": "index",
"title": "Reproduction audit",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1-theorem-3-3-literal-bound",
"title": "Claim 1: Theorem 3.3 proves that any k-layer state-space model solving the function-composition tasks under injectivity conditions must have total log state-space size scaling as \u03a9(m\u00b7log|V| \u2212 q\u00b7log|Y|), linear in the hidden dimension m (Theorem 3.3).",
"file": "pages/claim-1-theorem-3-3-literal-bound/page.md",
"children": []
},
{
"slug": "claim-2-theorem-3-7-window-bound",
"title": "Claim 2: Theorem 3.7 proves that sliding-window Transformers solving the same tasks under a local-sensitivity condition require total window size scaling with the context-dependency range R (Theorem 3.7).",
"file": "pages/claim-2-theorem-3-7-window-bound/page.md",
"children": []
},
{
"slug": "claim-3-theorem-4-3-selective-copy",
"title": "Claim 3: Theorem 4.3 constructs a two-layer hybrid (Mamba + attention) model that solves the selective copying task using embedding dimension O(max(log|V|, log L)) and working memory \u00d5(N), versus \u03a9(L) required by pure Transformers (Theorem 4.3).",
"file": "pages/claim-3-theorem-4-3-selective-copy/page.md",
"children": []
},
{
"slug": "claim-4-theorem-4-6-associative-recall",
"title": "Claim 4: Theorem 4.6 constructs a three-layer hybrid model that achieves 99% accuracy on the associative recall task using embedding dimension O(max(log|V|, log L)) and window size \u00d5(|V|) (Theorem 4.6).",
"file": "pages/claim-4-theorem-4-6-associative-recall/page.md",
"children": []
},
{
"slug": "claim-5-figure-4-selective-copy-learning",
"title": "Claim 5: On the selective copying task, the learned hybrid model reaches perfect accuracy with roughly 2,000 parameters while pure Transformer/SSM models need roughly 12,000 parameters to match it, a 6x parameter gap (Figure 4).",
"file": "pages/claim-5-figure-4-selective-copy-learning/page.md",
"children": []
},
{
"slug": "claim-6-figures-5-6-associative-recall-learning",
"title": "Claim 6: On multi-key associative recall, the hybrid model reaches 60% accuracy using 6x fewer parameters than pure Transformers, which plateau near 40% accuracy on single-key associative recall (Figures 5-6).",
"file": "pages/claim-6-figures-5-6-associative-recall-learning/page.md",
"children": []
}
]
},
"routes_built": {
"claim_pages": [
"pages/claim-1-theorem-3-3-literal-bound/page.md",
"pages/claim-2-theorem-3-7-window-bound/page.md",
"pages/claim-3-theorem-4-3-selective-copy/page.md",
"pages/claim-4-theorem-4-6-associative-recall/page.md",
"pages/claim-5-figure-4-selective-copy-learning/page.md",
"pages/claim-6-figures-5-6-associative-recall-learning/page.md"
]
}
}
|