{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:KCRU7TYZSHBOFDOHIYB74RMM5E","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"ab7b7c8573619592f3a7b9ec9deb8268e5e2cf91535a29a1df75056ca742f810","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T09:49:56Z","title_canon_sha256":"07c5f209588171790c38ceed67b05b60fde9b7ca4dc7724170c0db4ee2f5a181"},"schema_version":"1.0","source":{"id":"2605.13915","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13915","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13915v1","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13915","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"pith_short_12","alias_value":"KCRU7TYZSHBO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"KCRU7TYZSHBOFDOH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"KCRU7TYZ","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:2d5fc4677410fe1df89d033c1a9065424b0a56851872b3d86f33b5dff92973ea","target":"graph","created_at":"2026-05-17T23:39:18Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"For INT8 weights (W4A16), two-pass INT8 decomposition achieves near 16 effective bits. For MXFP4 weights (W4A16), two-pass MXFP4 decomposition yields near 6.6 effective bits with error bound 1/64 per block surpassing single-pass MXFP8 while maintaining the same effective GEMM compute time."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the multi-scale activation decomposition can be implemented with native hardware-accelerated GEMM without introducing pipeline stalls or accuracy loss beyond the derived bounds, and that the closed-form latency models accurately predict real hardware behavior."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MSD eliminates dequantization from the GEMM path by decomposing BF16 activations into multiple low-precision parts that multiply directly with INT8 or MXFP4 weights, achieving near-16 effective bits for INT8 and 6.6 for MXFP4 with reduced HBM traffic."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Decomposing BF16 activations into low-precision scales lets quantized weights multiply directly via native GEMM."}],"snapshot_sha256":"1c51d3ec24f85f8de4ab16604dd1578181b630d655d0ed344186bf5547b845af"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Quantization is essential for efficient large language model (LLM) inference, yet the dequantization step-converting low-bit weights back to high-precision for matrix multiplication has become a critical bottleneck on modern AI accelerators. On architectures with decoupled compute units (e.g., Ascend NPUs), dequantization operations can consume more cycles than the matrix multiplication itself, leaving the high-throughput tensor cores underutilized. This paper presents Multi-Scale Dequant (MSD), a quantization framework that removes weight/KV dequantization from the GEMM critical path. Instead","authors_text":"Chengqiu Hu, Fangzheng Miao, Jun Li, Junyi Fan, Lingchao Zheng, Qichen Liao, Rui Shi, Yuwei Fan","cross_cats":["cs.AI","cs.LG"],"headline":"Decomposing BF16 activations into low-precision scales lets quantized weights multiply directly via native GEMM.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T09:49:56Z","title":"Multi-Scale Dequant: Eliminating Dequantization Bottleneck via Activation Decomposition for Efficient LLM Inference"},"references":{"count":30,"internal_anchors":6,"resolved_work":30,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"DeepSeek.FlashMLA: Efficient MLA for Large Language Models. Technical Report, 2024. https://github.com/deepseek-ai/FlashMLA","work_id":"93f944c5-ef20-4968-bded-2807bc81f255","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Technical Blog, 2025.https://github.com/deepseek-ai/FlashMLA/blob/main/docs/ 20250929-hopper-fp8-sparse-deep-dive.md","work_id":"ab768edf-a287-47b5-ac22-b0b99458446e","year":2025},{"cited_arxiv_id":"2210.17323","doi":"","is_internal_anchor":true,"ref_index":3,"title":"GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers","work_id":"19ed8c44-202a-48f6-8169-637d5a5f2408","year":2022},{"cited_arxiv_id":"2306.00978","doi":"","is_internal_anchor":true,"ref_index":4,"title":"AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration","work_id":"ea9d1d72-db24-4cae-8c89-4ecd83dd87c1","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Castro, Jiale Chen, Torsten Hoefler, and Dan Alistarh","work_id":"19ce2d31-0683-4f81-9db6-b9ca47099921","year":2024}],"snapshot_sha256":"d39d51c469102fe8b184d2fbc89ed7dc8b3818489b20245e2d902f7caf5c1e76"},"source":{"id":"2605.13915","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T02:53:49.087136Z","id":"4b1150f9-d12f-4da5-a483-457377a4b710","model_set":{"reader":"grok-4.3"},"one_line_summary":"MSD eliminates dequantization from the GEMM path by decomposing BF16 activations into multiple low-precision parts that multiply directly with INT8 or MXFP4 weights, achieving near-16 effective bits for INT8 and 6.6 for MXFP4 with reduced HBM traffic.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Decomposing BF16 activations into low-precision scales lets quantized weights multiply directly via native GEMM.","strongest_claim":"For INT8 weights (W4A16), two-pass INT8 decomposition achieves near 16 effective bits. For MXFP4 weights (W4A16), two-pass MXFP4 decomposition yields near 6.6 effective bits with error bound 1/64 per block surpassing single-pass MXFP8 while maintaining the same effective GEMM compute time.","weakest_assumption":"That the multi-scale activation decomposition can be implemented with native hardware-accelerated GEMM without introducing pipeline stalls or accuracy loss beyond the derived bounds, and that the closed-form latency models accurately predict real hardware behavior."}},"verdict_id":"4b1150f9-d12f-4da5-a483-457377a4b710"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c4ec70c2bb33aad2bf75d9a76d12e95d10e8e8ca77f2f9001052724b445e73d3","target":"record","created_at":"2026-05-17T23:39:18Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"ab7b7c8573619592f3a7b9ec9deb8268e5e2cf91535a29a1df75056ca742f810","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T09:49:56Z","title_canon_sha256":"07c5f209588171790c38ceed67b05b60fde9b7ca4dc7724170c0db4ee2f5a181"},"schema_version":"1.0","source":{"id":"2605.13915","kind":"arxiv","version":1}},"canonical_sha256":"50a34fcf1991c2e28dc74603fe458ce90fbb394a604a1a63b5a39ad28dfaed34","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"50a34fcf1991c2e28dc74603fe458ce90fbb394a604a1a63b5a39ad28dfaed34","first_computed_at":"2026-05-17T23:39:18.763747Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:18.763747Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/SZgaWUJX6aMA8GuDOyZPjntzSsnfoZbXR8BRLjjBnJPN2bJuTCtnLhgDBKqk8NdjXUXI3VwGeaIBFms8fmvBA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:18.764477Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13915","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c4ec70c2bb33aad2bf75d9a76d12e95d10e8e8ca77f2f9001052724b445e73d3","sha256:2d5fc4677410fe1df89d033c1a9065424b0a56851872b3d86f33b5dff92973ea"],"state_sha256":"2ad7d04e34583283d11bd3a9bb79de74d408d57a3382556c975a2d17365cd33e"}