{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:PBIHCRXCYDLTJOT2QLT3VN5R6C","short_pith_number":"pith:PBIHCRXC","canonical_record":{"source":{"id":"2603.29002","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-03-30T21:03:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0172be3fd73ca1eda9a2e50165458f25eabed54f9744e01cc3b99cc29b7e3360","abstract_canon_sha256":"67e180a5f6292f7cdb7caaf22cfc357bb5e04f73a0c942d9a401f8e3f8ddc65d"},"schema_version":"1.0"},"canonical_sha256":"78507146e2c0d734ba7a82e7bab7b1f088de63fc3e1709cf2bfcf66309130942","source":{"kind":"arxiv","id":"2603.29002","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.29002","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"arxiv_version","alias_value":"2603.29002v3","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.29002","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_12","alias_value":"PBIHCRXCYDLT","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_16","alias_value":"PBIHCRXCYDLTJOT2","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_8","alias_value":"PBIHCRXC","created_at":"2026-06-02T01:03:46Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:PBIHCRXCYDLTJOT2QLT3VN5R6C","target":"record","payload":{"canonical_record":{"source":{"id":"2603.29002","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-03-30T21:03:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0172be3fd73ca1eda9a2e50165458f25eabed54f9744e01cc3b99cc29b7e3360","abstract_canon_sha256":"67e180a5f6292f7cdb7caaf22cfc357bb5e04f73a0c942d9a401f8e3f8ddc65d"},"schema_version":"1.0"},"canonical_sha256":"78507146e2c0d734ba7a82e7bab7b1f088de63fc3e1709cf2bfcf66309130942","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:46.223302Z","signature_b64":"ScqD9xbgEp/FynzQmXwmD8xIx3Y3wySfx3zbGNlv4kUsMwf/+dYxlCDZSaDEhk2r7PEfHCTqYzohcGkB6zPZCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"78507146e2c0d734ba7a82e7bab7b1f088de63fc3e1709cf2bfcf66309130942","last_reissued_at":"2026-06-02T01:03:46.222739Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:46.222739Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2603.29002","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T01:03:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"A5bBvXOkeKgPAYR8iDvZ8A0oZd6NB37yY373q1ocbFUQa+BdJUS+7eB3hwk1tRxst+ce3uyoXN8Jio9VCuy+BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T07:50:09.382415Z"},"content_sha256":"dd34d4910271e4f0f6c8642c8fead6a8292fd610242d55742e4648641e5b2b37","schema_version":"1.0","event_id":"sha256:dd34d4910271e4f0f6c8642c8fead6a8292fd610242d55742e4648641e5b2b37"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:PBIHCRXCYDLTJOT2QLT3VN5R6C","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Understand and Accelerate Memory Processing Pipeline for Large Language Model Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations.","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Jason Cong, Rui Ma, Yizhou Sun, Zifan He","submitted_at":"2026-03-30T21:03:39Z","abstract_excerpt":"Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse attention, retrieval-augmented generation (RAG), and compressed contextual memory, to support complex reasoning. We show that these optimizations can be unified into a four-step memory processing pipeline: Prepare Memory, Compute Relevancy, Retrieval, and Apply to Inference. Through systematic profiling, we identify a 22%-97% memory processing overhead in LLM inference and strong heterogeneity in its computational characteristics. Motivated by this insight, "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We demonstrate this approach on a GPU-FPGA system by offloading sparse, irregular, and memory-bounded operations to FPGAs while retaining compute-intensive operations on GPUs. Evaluated on an AMD MI210 GPU and an Alveo U55C FPGA, our system is up to 2.2× faster and achieves up to 4.7× less energy across multiple LLM inference optimizations than the GPU baseline.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The profiling results and workload heterogeneity observed on the tested models and inputs will generalize to other LLMs, and that FPGA offloading incurs negligible data-transfer overheads relative to the gains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Unifying LLM memory optimizations into a Prepare-Compute-Retrieve-Apply pipeline and accelerating it on GPU-FPGA hardware yields up to 2.2x faster inference and 4.7x less energy than GPU-only baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7dab31d1bf094392ae188d25bf841ca2b6a53792be7d8adc7c87d7f475571899"},"source":{"id":"2603.29002","kind":"arxiv","version":3},"verdict":{"id":"13a31856-b4c8-4075-b7ae-4e766754d753","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T00:22:39.912975Z","strongest_claim":"We demonstrate this approach on a GPU-FPGA system by offloading sparse, irregular, and memory-bounded operations to FPGAs while retaining compute-intensive operations on GPUs. Evaluated on an AMD MI210 GPU and an Alveo U55C FPGA, our system is up to 2.2× faster and achieves up to 4.7× less energy across multiple LLM inference optimizations than the GPU baseline.","one_line_summary":"Unifying LLM memory optimizations into a Prepare-Compute-Retrieve-Apply pipeline and accelerating it on GPU-FPGA hardware yields up to 2.2x faster inference and 4.7x less energy than GPU-only baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The profiling results and workload heterogeneity observed on the tested models and inputs will generalize to other LLMs, and that FPGA offloading incurs negligible data-transfer overheads relative to the gains.","pith_extraction_headline":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.29002/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"13a31856-b4c8-4075-b7ae-4e766754d753"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T01:03:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Clw0nLottTi6t6xi/pIoC0EH9T3H6So7/SdDMyIc7albGRrf3Pq/5S8t23zkuFwAq9verk7cqLYpboNk2fDqBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T07:50:09.382895Z"},"content_sha256":"c2958500e1aecdfb993b5987eb3f039fb66addedd181681d8dbfe100e27fa625","schema_version":"1.0","event_id":"sha256:c2958500e1aecdfb993b5987eb3f039fb66addedd181681d8dbfe100e27fa625"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/bundle.json","state_url":"https://pith.science/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-02T07:50:09Z","links":{"resolver":"https://pith.science/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C","bundle":"https://pith.science/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/bundle.json","state":"https://pith.science/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/state.json","well_known_bundle":"https://pith.science/.well-known/pith/PBIHCRXCYDLTJOT2QLT3VN5R6C/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:PBIHCRXCYDLTJOT2QLT3VN5R6C","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":"67e180a5f6292f7cdb7caaf22cfc357bb5e04f73a0c942d9a401f8e3f8ddc65d","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-03-30T21:03:39Z","title_canon_sha256":"0172be3fd73ca1eda9a2e50165458f25eabed54f9744e01cc3b99cc29b7e3360"},"schema_version":"1.0","source":{"id":"2603.29002","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.29002","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"arxiv_version","alias_value":"2603.29002v3","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.29002","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_12","alias_value":"PBIHCRXCYDLT","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_16","alias_value":"PBIHCRXCYDLTJOT2","created_at":"2026-06-02T01:03:46Z"},{"alias_kind":"pith_short_8","alias_value":"PBIHCRXC","created_at":"2026-06-02T01:03:46Z"}],"graph_snapshots":[{"event_id":"sha256:c2958500e1aecdfb993b5987eb3f039fb66addedd181681d8dbfe100e27fa625","target":"graph","created_at":"2026-06-02T01:03:46Z","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":"We demonstrate this approach on a GPU-FPGA system by offloading sparse, irregular, and memory-bounded operations to FPGAs while retaining compute-intensive operations on GPUs. Evaluated on an AMD MI210 GPU and an Alveo U55C FPGA, our system is up to 2.2× faster and achieves up to 4.7× less energy across multiple LLM inference optimizations than the GPU baseline."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The profiling results and workload heterogeneity observed on the tested models and inputs will generalize to other LLMs, and that FPGA offloading incurs negligible data-transfer overheads relative to the gains."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Unifying LLM memory optimizations into a Prepare-Compute-Retrieve-Apply pipeline and accelerating it on GPU-FPGA hardware yields up to 2.2x faster inference and 4.7x less energy than GPU-only baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations."}],"snapshot_sha256":"7dab31d1bf094392ae188d25bf841ca2b6a53792be7d8adc7c87d7f475571899"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.29002/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse attention, retrieval-augmented generation (RAG), and compressed contextual memory, to support complex reasoning. We show that these optimizations can be unified into a four-step memory processing pipeline: Prepare Memory, Compute Relevancy, Retrieval, and Apply to Inference. Through systematic profiling, we identify a 22%-97% memory processing overhead in LLM inference and strong heterogeneity in its computational characteristics. Motivated by this insight, ","authors_text":"Jason Cong, Rui Ma, Yizhou Sun, Zifan He","cross_cats":["cs.AI"],"headline":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-03-30T21:03:39Z","title":"Understand and Accelerate Memory Processing Pipeline for Large Language Model Inference"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.29002","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-14T00:22:39.912975Z","id":"13a31856-b4c8-4075-b7ae-4e766754d753","model_set":{"reader":"grok-4.3"},"one_line_summary":"Unifying LLM memory optimizations into a Prepare-Compute-Retrieve-Apply pipeline and accelerating it on GPU-FPGA hardware yields up to 2.2x faster inference and 4.7x less energy than GPU-only baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Heterogeneous GPU-FPGA systems accelerate LLM memory processing up to 2.2x by offloading sparse operations.","strongest_claim":"We demonstrate this approach on a GPU-FPGA system by offloading sparse, irregular, and memory-bounded operations to FPGAs while retaining compute-intensive operations on GPUs. Evaluated on an AMD MI210 GPU and an Alveo U55C FPGA, our system is up to 2.2× faster and achieves up to 4.7× less energy across multiple LLM inference optimizations than the GPU baseline.","weakest_assumption":"The profiling results and workload heterogeneity observed on the tested models and inputs will generalize to other LLMs, and that FPGA offloading incurs negligible data-transfer overheads relative to the gains."}},"verdict_id":"13a31856-b4c8-4075-b7ae-4e766754d753"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:dd34d4910271e4f0f6c8642c8fead6a8292fd610242d55742e4648641e5b2b37","target":"record","created_at":"2026-06-02T01:03:46Z","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":"67e180a5f6292f7cdb7caaf22cfc357bb5e04f73a0c942d9a401f8e3f8ddc65d","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-03-30T21:03:39Z","title_canon_sha256":"0172be3fd73ca1eda9a2e50165458f25eabed54f9744e01cc3b99cc29b7e3360"},"schema_version":"1.0","source":{"id":"2603.29002","kind":"arxiv","version":3}},"canonical_sha256":"78507146e2c0d734ba7a82e7bab7b1f088de63fc3e1709cf2bfcf66309130942","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"78507146e2c0d734ba7a82e7bab7b1f088de63fc3e1709cf2bfcf66309130942","first_computed_at":"2026-06-02T01:03:46.222739Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-02T01:03:46.222739Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ScqD9xbgEp/FynzQmXwmD8xIx3Y3wySfx3zbGNlv4kUsMwf/+dYxlCDZSaDEhk2r7PEfHCTqYzohcGkB6zPZCg==","signature_status":"signed_v1","signed_at":"2026-06-02T01:03:46.223302Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.29002","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:dd34d4910271e4f0f6c8642c8fead6a8292fd610242d55742e4648641e5b2b37","sha256:c2958500e1aecdfb993b5987eb3f039fb66addedd181681d8dbfe100e27fa625"],"state_sha256":"85ef6cfc919174c3c7fa706f389f70a0812e164651b1689e237aaf8e57433bd3"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"z9QyZODBfaxQx3UXAR0zb8GWEdbR63rq9axa5ty72nhoAQmlczq/vkXjJYIIi1ADmWnGdOsf5FDHN0eA0CXaBw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-02T07:50:09.385165Z","bundle_sha256":"8f5364ad0450aa918e53e3314fb1bfc4bfb49d81289ba653cc881daf69d54472"}}