{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/code/initialize-past-key-values","entry":"initialize_past_key_values","source":"Syntology graph, per-sample; not an archive number","read_at":"2026-09-24T18:15:14+00:00","claim":"Names are grouped by exact entry-name string. Same-named routines are NOT asserted to be equivalent; 'ran' means executed on a synthesized fixture, not correctness. n_samples_ran = sum of by_status over every status except 'unverified' (ran_draft_wrong and ran_fixture are failures of Syntology's instrument, not of the code); n_papers_ran = papers with at least one such sample.","status_vocabulary":{"ran_honours":"ran, honoured the contract we drafted","ran_violates":"ran, violated the contract we drafted","ran_draft_wrong":"ran; our contract draft was wrong, not the code","ran_fixture":"ran; our fixture could not drive it","ran":"ran on a synthesized input","unverified":"unverified (harvested, no recorded run)"},"n_papers":15,"n_papers_ran":10,"units":"n_samples, n_samples_ran, n_samples_fingerprinted and by_status count distinct code bodies (code_sha256); n_places and n_places_pointer_only count places, one per (paper, code body) pair, which is also the unit of the samples list","n_samples":10,"n_samples_ran":6,"n_samples_fingerprinted":0,"n_places":15,"n_places_pointer_only":4,"by_status":{"ran_honours":0,"ran_violates":0,"ran_draft_wrong":3,"ran_fixture":0,"ran":3,"unverified":4},"syntology":{"atlas_url":null,"mcp":null,"mcp_per_sample":{"tool":"get_code","arguments_in":"samples[].mcp_get_code"},"developers":"https://syntology.ai/developers"},"samples":[{"arxiv_id":"2606.00535","paper":"/paper/arxiv-2606-00535","title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","date":null,"month_inferred_from_arxiv_id":"2026-06","title_source":"syntology","repo":"SAI-Lab-NYU/DREAM-S","path":"dream_s/model/kv_cache.py","file_url":"https://github.com/SAI-Lab-NYU/DREAM-S/blob/HEAD/dream_s/model/kv_cache.py","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"code_sha256_prefix":"2f5e7feccee982b7","mcp_get_code":{"code_sha256":"2f5e7feccee982b7"}},{"arxiv_id":"2602.16052","paper":"/paper/arxiv-2602-16052","title":"MoE-Spec: Expert Budgeting for Efficient Speculative Decoding","date":null,"month_inferred_from_arxiv_id":"2026-02","title_source":"syntology","repo":"SafeAILab/EAGLE","path":"eagle/model/kv_cache.py","file_url":"https://github.com/SafeAILab/EAGLE/blob/HEAD/eagle/model/kv_cache.py","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NOASSERTION","inline_ok":false,"code_sha256_prefix":"bbfbabefe7ad3955","mcp_get_code":{"code_sha256":"bbfbabefe7ad3955"}},{"arxiv_id":"2601.19278","paper":"/paper/arxiv-2601-19278","title":"DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference","date":null,"month_inferred_from_arxiv_id":"2026-01","title_source":"syntology","repo":"fvliang/DART","path":"dart/model/kv_cache.py","file_url":"https://github.com/fvliang/DART/blob/HEAD/dart/model/kv_cache.py","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"9315d4be39a16b19","mcp_get_code":{"code_sha256":"9315d4be39a16b19"}},{"arxiv_id":"2601.15498","paper":"/paper/arxiv-2601-15498","title":"MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification","date":null,"month_inferred_from_arxiv_id":"2026-01","title_source":"syntology","repo":"5SSjw/MARS","path":"mars/model/kv_cache.py","file_url":"https://github.com/5SSjw/MARS/blob/HEAD/mars/model/kv_cache.py","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"code_sha256_prefix":"bbfbabefe7ad3955","mcp_get_code":{"code_sha256":"bbfbabefe7ad3955"}},{"arxiv_id":"2509.15235","paper":"/paper/arxiv-2509-15235","title":"ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding","date":null,"month_inferred_from_arxiv_id":"2025-09","title_source":"syntology","repo":"KangJialiang/ViSpec","path":"vispec/model/kv_cache.py","file_url":"https://github.com/KangJialiang/ViSpec/blob/HEAD/vispec/model/kv_cache.py","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"86a9a36211cc9bb5","mcp_get_code":{"code_sha256":"86a9a36211cc9bb5"}},{"arxiv_id":"2508.16201","paper":"/paper/arxiv-2508-16201","title":"SPECVLM: Enhancing Speculative Decoding of Video LLMs via Verifier-Guided Token Pruning","date":null,"month_inferred_from_arxiv_id":"2025-08","title_source":"syntology","repo":"zju-jiyicheng/SpecVLM","path":"decoding/tree_decoding.py","file_url":"https://github.com/zju-jiyicheng/SpecVLM/blob/HEAD/decoding/tree_decoding.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"code_sha256_prefix":"d992d12cd63a20e0","mcp_get_code":{"code_sha256":"d992d12cd63a20e0"}},{"arxiv_id":"2507.22424","paper":null,"title":"arXiv:2507.22424","date":null,"month_inferred_from_arxiv_id":"2025-07","title_source":null,"repo":"PineTreeWss/SpecVLA","path":"openvla/specdecoding/model/kv_cache.py","file_url":"https://github.com/PineTreeWss/SpecVLA/blob/HEAD/openvla/specdecoding/model/kv_cache.py","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"code_sha256_prefix":"2fa0d32bf1353ed5","mcp_get_code":{"code_sha256":"2fa0d32bf1353ed5"}},{"arxiv_id":"2503.10135","paper":"/paper/gumiho-a-hybrid-architecture-to-prioritize","title":"Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding","date":"2025-03-13","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"AMD-AIG-AIMA/Gumiho","path":"gumiho/model/kv_cache.py","file_url":"https://github.com/AMD-AIG-AIMA/Gumiho/blob/HEAD/gumiho/model/kv_cache.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"aadc9c83d7ec2701","mcp_get_code":{"code_sha256":"aadc9c83d7ec2701"}},{"arxiv_id":"2410.06916","paper":"/paper/swift-on-the-fly-self-speculative-decoding","title":"SWIFT: On-the-Fly Self-Speculative Decoding for LLM Inference Acceleration","date":"2024-10-09","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"hemingkx/SWIFT","path":"evaluation_llama/inference_swift.py","file_url":"https://github.com/hemingkx/SWIFT/blob/HEAD/evaluation_llama/inference_swift.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"aadc9c83d7ec2701","mcp_get_code":{"code_sha256":"aadc9c83d7ec2701"}},{"arxiv_id":"2408.00264","paper":"/paper/clover-2-accurate-inference-for-regressive","title":"Clover-2: Accurate Inference for Regressive Lightweight Speculative Decoding","date":"2024-08-01","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"XiaoBin1992/clover","path":"clover/model/kv_cache.py","file_url":"https://github.com/XiaoBin1992/clover/blob/HEAD/clover/model/kv_cache.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"aadc9c83d7ec2701","mcp_get_code":{"code_sha256":"aadc9c83d7ec2701"}},{"arxiv_id":"2406.17404","paper":"/paper/make-some-noise-unlocking-language-model","title":"Make Some Noise: Unlocking Language Model Parallel Inference Capability through Noisy Training","date":"2024-06-25","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"wyxstriker/MakeSomeNoiseInference","path":"model/noise_forward.py","file_url":"https://github.com/wyxstriker/MakeSomeNoiseInference/blob/HEAD/model/noise_forward.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"code_sha256_prefix":"56a9074f8ed59e83","mcp_get_code":{"code_sha256":"56a9074f8ed59e83"}},{"arxiv_id":"2402.05109","paper":"/paper/hydra-sequentially-dependent-draft-heads-for","title":"Hydra: Sequentially-Dependent Draft Heads for Medusa Decoding","date":"2024-02-07","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"zankner/hydra","path":"hydra/model/kv_cache.py","file_url":"https://github.com/zankner/hydra/blob/HEAD/hydra/model/kv_cache.py","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"d2d232761fe51ccb","mcp_get_code":{"code_sha256":"d2d232761fe51ccb"}},{"arxiv_id":"2401.10774","paper":"/paper/medusa-simple-llm-inference-acceleration","title":"Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads","date":"2024-01-19","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"fasterdecoding/medusa","path":"medusa/model/kv_cache.py","file_url":"https://github.com/fasterdecoding/medusa/blob/HEAD/medusa/model/kv_cache.py","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"56a9074f8ed59e83","mcp_get_code":{"code_sha256":"56a9074f8ed59e83"}},{"arxiv_id":"2311.08252","paper":"/paper/rest-retrieval-based-speculative-decoding","title":"REST: Retrieval-Based Speculative Decoding","date":"2023-11-14","month_inferred_from_arxiv_id":null,"title_source":"archive","repo":"fasterdecoding/rest","path":"rest/model/kv_cache.py","file_url":"https://github.com/fasterdecoding/rest/blob/HEAD/rest/model/kv_cache.py","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"3965aca677b30513","mcp_get_code":{"code_sha256":"3965aca677b30513"}},{"arxiv_id":"2024.naacl-long.88","paper":null,"title":"arXiv:2024.naacl-long.88","date":null,"month_inferred_from_arxiv_id":null,"title_source":null,"repo":"FasterDecoding/REST","path":"rest/model/kv_cache.py","file_url":"https://github.com/FasterDecoding/REST/blob/HEAD/rest/model/kv_cache.py","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"code_sha256_prefix":"3965aca677b30513","mcp_get_code":{"code_sha256":"3965aca677b30513"}}]}