{
  "schema_version": 1,
  "title": "Reproduction: Training–Inference Consistent Segmented Execution for Long-Context LLMs",
  "emoji": "🎯",
  "space_id": "Umong/repro-training-inference-consistent-segmented-execution-for-long-context-llms",
  "paper": {
    "arxiv_id": "2605.11744"
  },
  "tags": [
    "icml2026-repro",
    "paper-PoRigyDOcC"
  ],
  "updated_at": "2026-07-18T18:43:55+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: Training–Inference Consistent Segmented Execution for Long-Context LLMs",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-the-framework-defines-segment-level-execution-semantics-where-training-and-inference-process-sequences-segment-by-segment-with-the-same-cross-segment-interface-definition-3-1",
        "title": "Claim 1: The framework defines segment-level execution semantics where training and inference process sequences segment by segment with the same cross-segment interface (Definition 3.1)",
        "file": "pages/claim-1-the-framework-defines-segment-level-execution-semantics-where-training-and-inference-process-sequences-segment-by-segment-with-the-same-cross-segment-interface-definition-3-1/page.md",
        "children": []
      },
      {
        "slug": "claim-2-for-the-stated-truncated-consistent-objective-tbptt-computes-the-exact-gradient-rather-than-an-approximation-theorem-3-3",
        "title": "Claim 2: For the stated truncated consistent objective, TBPTT computes the exact gradient rather than an approximation (Theorem 3.3)",
        "file": "pages/claim-2-for-the-stated-truncated-consistent-objective-tbptt-computes-the-exact-gradient-rather-than-an-approximation-theorem-3-3/page.md",
        "children": []
      },
      {
        "slug": "claim-3-training-inference-alignment-follows-when-the-same-segmented-execution-semantics-and-truncated-objective-are-used-for-training-and-inference-corollary-3-4",
        "title": "Claim 3: Training-inference alignment follows when the same segmented execution semantics and truncated objective are used for training and inference (Corollary 3.4)",
        "file": "pages/claim-3-training-inference-alignment-follows-when-the-same-segmented-execution-semantics-and-truncated-objective-are-used-for-training-and-inference-corollary-3-4/page.md",
        "children": []
      },
      {
        "slug": "claim-4-the-architecture-uses-head-and-layer-sparse-long-range-retrieval-with-carried-kv-tails-and-forward-only-retrieved-prefixes-figure-3",
        "title": "Claim 4: The architecture uses head- and layer-sparse long-range retrieval with carried KV tails and forward-only retrieved prefixes (Figure 3)",
        "file": "pages/claim-4-the-architecture-uses-head-and-layer-sparse-long-range-retrieval-with-carried-kv-tails-and-forward-only-retrieved-prefixes-figure-3/page.md",
        "children": []
      },
      {
        "slug": "claim-5-the-method-achieves-comparable-longbench-e-performance-while-lowering-prefill-memory-and-latency-relative-to-full-context-attention-and-other-efficient-baselines-table-1",
        "title": "Claim 5: The method achieves comparable LongBench-E performance while lowering prefill memory and latency relative to full-context attention and other efficient baselines (Table 1)",
        "file": "pages/claim-5-the-method-achieves-comparable-longbench-e-performance-while-lowering-prefill-memory-and-latency-relative-to-full-context-attention-and-other-efficient-baselines-table-1/page.md",
        "children": []
      },
      {
        "slug": "claim-6-at-128k-context-segmented-execution-provides-approximately-6x-lower-peak-prefill-memory-than-full-context-attention-with-flashattention-figure-5",
        "title": "Claim 6: At 128K context, segmented execution provides approximately 6x lower peak prefill memory than full-context attention with FlashAttention (Figure 5)",
        "file": "pages/claim-6-at-128k-context-segmented-execution-provides-approximately-6x-lower-peak-prefill-memory-than-full-context-attention-with-flashattention-figure-5/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  },
  "agent_view_tokens": 4151,
  "revision": "1784400235352425480"
}