Hongseok Kim, Minsoo Kim, Jungwook Choi, Hyungmin Kim
12 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 7 failures of our instrument; 5 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.
"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2606.06302.json · badge: /api/ran/2606.06302.svg
Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.
| Repository | Role | Ran |
|---|---|---|
| aiha-lab/TANGRAM | canonical (classifier label) classifier basis: census |
12 ran · 5 no recorded run |
| Function | Status | Where it lives |
|---|---|---|
| async_request_deepspeed_mii | Ran ran on a synthesized input |
aiha-lab/TANGRAM/benchmarks/backend_request_func.py pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("552e4370076a1a01") How it ranThe body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| async_request_tgi | Ran ran on a synthesized input |
aiha-lab/TANGRAM/benchmarks/backend_request_func.py pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("7ee13b8d142453c0") How it ranThe body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| async_request_trt_llm | Ran ran on a synthesized input |
aiha-lab/TANGRAM/benchmarks/backend_request_func.py pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("f0b0f51ad4ed00c9") How it ranThe body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| build_prompt | Ran ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT) |
aiha-lab/TANGRAM/benchmarks/tangram/ruler_local.py code served (permissive licence) · get_code("bcd9d69939662364") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| effective_ratio | Ran ran, honoured the contract we drafted |
aiha-lab/TANGRAM/benchmarks/tangram/bench_common.py code served (permissive licence) · get_code("4d80b1ddfa08469e") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| extract_request_timing | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/bench_common.py code served (permissive licence) · get_code("418bfc593ad8d8bf") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| get_data_list | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/scbench_local.py code served (permissive licence) · get_code("550815cfa153d37c") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| get_eval_task | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/benchmark_scbench.py code served (permissive licence) · get_code("6439d722e41feedf") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| get_query_text | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/benchmark_scbench.py code served (permissive licence) · get_code("941d33a97c872155") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| group_by_task | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/ruler_local.py code served (permissive licence) · get_code("d59e2d450cbbf7a4") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| set_gen_length | Ran ran, honoured the contract we drafted |
aiha-lab/TANGRAM/benchmarks/tangram/scbench_local.py code served (permissive licence) · get_code("ac80359234193760") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| template | Ran ran; our contract draft was wrong, not the code |
aiha-lab/TANGRAM/benchmarks/tangram/scbench_local.py code served (permissive licence) · get_code("33321941cb7c5a2a") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
|
| load_ruler | Not yet run unverified (harvested, no recorded run) |
aiha-lab/TANGRAM/benchmarks/tangram/ruler_local.py code served (permissive licence) · get_code("42a1756632040233") How it ranno invocation recorded. |
| resolve_cluster_map | Not yet run unverified (harvested, no recorded run) |
aiha-lab/TANGRAM/benchmarks/tangram/speedup/speedup_benchmark.py code served (permissive licence) · get_code("f942346fd9105175") How it ranno invocation recorded. |
| run_config_in_subprocess | Not yet run unverified (harvested, no recorded run) |
aiha-lab/TANGRAM/benchmarks/tangram/preemption_overhead/benchmark.py code served (permissive licence) · get_code("db377885d2303bb9") How it ranno invocation recorded. |
| run_one | Not yet run unverified (harvested, no recorded run) |
aiha-lab/TANGRAM/benchmarks/tangram/preemption_overhead/benchmark.py code served (permissive licence) · get_code("2298f1ff53dfe660") How it ranno invocation recorded. |
| run_one | Not yet run unverified (harvested, no recorded run) |
aiha-lab/TANGRAM/benchmarks/tangram/speedup/speedup_benchmark.py code served (permissive licence) · get_code("8fb5af6a89b21e9b") How it ranno invocation recorded. |
Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 3 later) and 14 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).
Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.
| Function | Also present in | Earliest-dated | Others named |
|---|---|---|---|
| async_request_deepspeed_mii | 2 other papers 1 dated earlier, 1 later |
2602.07223 (2026-02) Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention |
2608.20359 (2026-08) Self-Speculation for Faster Reasoning Models |
| async_request_tgi | 2 other papers 1 dated earlier, 1 later |
2602.07223 (2026-02) Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention |
2608.20359 (2026-08) Self-Speculation for Faster Reasoning Models |
| async_request_trt_llm | 2 other papers 1 dated earlier, 1 later |
2602.07223 (2026-02) Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention |
2608.20359 (2026-08) Self-Speculation for Faster Reasoning Models |
Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Multi-turn LLM serving accumulates dialogue history whose Key-Value (KV) cache grows with every turn and every user, quickly exceeding the model weights themselves and making memory-not compute-the binding constraint on throughput. Non-uniform KV compression, which allocates heterogeneous budgets across attention heads, preserves accuracy far better than uniform schemes, yet remains impractical: modern serving stacks assume identical KV lengths across heads, so heterogeneity traps freed memory as page fragmentation, spends up to 25% of prefill time reclaiming scattered pages, and skews GPU workloads that inflate decode latency by up to 1.7× or burn 15-20% of each decode step on re-planning. We observe that this heterogeneity need not be discovered at runtime: head-wise retention follows a two-level structural regularity-an input-invariant head ranking with narrowly bounded per-head ratios-that can be calibrated offline from as few as 50 samples. Building on this insight, we present Tangram, a serving framework that statically resolves what prior systems handle dynamically: Budget Reservation fixes each head's post-compression footprint at scheduling time, eliminating page reclamation; Ragged Paging clusters similar-budget heads into independent page tables, turning fragmentation into reclaimable memory; and Ahead-of-Time Load Balancing precomputes balanced GPU partitions with zero runtime planning. Implemented on vLLM, Tangram serves as a drop-in substrate for existing non-uniform compression methods, matching their accuracy while improving end-to-end throughput by up to 2.6× over the full-KV baseline. Our implementation is publicly available at https://github.com/aiha-lab/TANGRAM. * Equal contribution † Currently at Apple ‡ Corresponding author consistent, personalized responses [1,3,24,28]. To avoid recomputing this ever-growing history H 𝑡 at every turn [12,20,26,37], serving systems persist attention states in the Key-Value (KV) cache [32]-but the cache grows linearly with every turn and every concurrent user. For Qwen2.5-32B, with merely 16 concurrent requests, the accumulated KV cache surpasses the size of the model weights themselves within ten conversation turns, and continues to grow unboundedly thereafter (Figure 1(a)). In multi-turn serving, memory capacity-not compute-is the binding constraint on batch size and therefore on throughput. KV cache compression is the standard remedy, but how the token budget is distributed across attention heads determines whether accuracy survives. Uniform compression [14,18,22,29,40,45] forces every head to retain the same number of tokens, ignoring a fundamental property of attention: critical long-range information is concentrated in a small subset of retrieval heads [10], while other heads attend only locally. Truncating all heads equally starves precisely the heads that matter, and accuracy degrades sharply in multi-turn settings (Figure 1(c)). 10,16,34,39] instead allocates heterogeneous per-head budgets that mirror this skew, preserving near-original accuracy even under aggressive compression. Thus, non-uniform compression is the right tool for memory-efficient multi-turn serving. Systemically, however, non-uniform compression is impractical today. The software stack of state-of-the-art serving systems-PagedAttention [19], continuous batching with chunked prefill [2, 42], and optimized attention kernels [5,6,41]-is architected end-to-end around a single implicit assumption: all attention heads hold KV caches of identical length. Non-uniform compression violates this assumption at every layer of the stack, resulting in severe overhead. First, the monolithic page of PagedAttention [19] spans all heads at a single uniform length, so the heterogeneous per-head budgets of non-uniform compression cannot be realized on it in the first place; the memory that compression would free instead stays trapped as unrecoverable page fragmentation ( § 3.1.1). Second, because each head's post-compression 1
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2606.06302")
get_code_for_paper("2606.06302")
have("2606.06302")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2606.06302.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2606.06302)
Connect an agent — have() is free.