{"paper":{"title":"Graph Memory Transformer (GMT)","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A learned graph of memory centroids can replace the feed-forward sublayer in a decoder-only transformer.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Niccol\\`o Ferrari, Nicola Zanarini","submitted_at":"2026-04-26T20:09:25Z","abstract_excerpt":"We investigate whether the Feed-Forward Network (FFN) sublayer in a decoder-only transformer can be replaced by an explicit learned memory graph while preserving the surrounding autoregressive architecture. The proposed Graph Memory Transformer (GMT) keeps causal self-attention intact, but replaces the usual per-token FFN transformation with a memory cell that routes token representations over a learned bank of centroids connected by a learned directed transition matrix. In the base GMT v7 instantiation studied here, each of 16 transformer blocks contains 128 centroids, a 128 * 128 edge matrix"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The base GMT v7 model trains stably and exposes centroid usage, transition structure, and source-to-target movement as directly inspectable quantities of the forward computation. It remains behind the larger dense baseline in validation loss and perplexity (3.5995/36.58 vs. 3.2903/26.85), while showing close zero-shot benchmark behavior under the evaluated setting.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That gravitational source routing plus token-conditioned target selection plus gated displacement readout can functionally substitute for the dense FFN transformation without requiring additional capacity or architectural changes outside the memory cell itself.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Graph Memory Transformer (GMT) swaps dense FFN sublayers for a graph of 128 centroids and a learned 128x128 transition matrix per block, yielding a 82M-parameter decoder-only LM that trains stably but trails a 103M dense baseline in perplexity.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A learned graph of memory centroids can replace the feed-forward sublayer in a decoder-only transformer.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"622e7f307e16a4a43f6e050b20f4b25425ce585f5d9c7e61d1f9238ecdf6a960"},"source":{"id":"2604.23862","kind":"arxiv","version":2},"verdict":{"id":"66cf2d24-7e13-42cf-bc27-690c491710d1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T06:24:14.512096Z","strongest_claim":"The base GMT v7 model trains stably and exposes centroid usage, transition structure, and source-to-target movement as directly inspectable quantities of the forward computation. It remains behind the larger dense baseline in validation loss and perplexity (3.5995/36.58 vs. 3.2903/26.85), while showing close zero-shot benchmark behavior under the evaluated setting.","one_line_summary":"Graph Memory Transformer (GMT) swaps dense FFN sublayers for a graph of 128 centroids and a learned 128x128 transition matrix per block, yielding a 82M-parameter decoder-only LM that trains stably but trails a 103M dense baseline in perplexity.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That gravitational source routing plus token-conditioned target selection plus gated displacement readout can functionally substitute for the dense FFN transformation without requiring additional capacity or architectural changes outside the memory cell itself.","pith_extraction_headline":"A learned graph of memory centroids can replace the feed-forward sublayer in a decoder-only transformer."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.23862/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-21T07:43:11.410025Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T22:42:36.913523Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"f0922dd35a64782136d1f08b38c0aeb566f919a7180a5d3b6d3866065d4fb412"},"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"}