{"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":"/paper/arxiv-2603-21461","title":"DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment","arxiv_id":"2603.21461","date":"2026-03-23","proceeding":null,"authors":["James Wedgwood","Aashiq Muhamed","Mona T. Diab","Virginia Smith"],"abstract":"Preference alignment is usually achieved by weight-updating training on preference data, which adds substantial alignment-stage compute and provides limited mechanistic visibility. We propose Dynamic SAE Steering for Preference Alignment (DSPA), an inference-time method that makes sparse autoencoder (SAE) steering prompt-conditional. From preference triples, DSPA computes a conditional-difference map linking prompt features to generation-control features; during decoding, it modifies only token-active latents, without base-model weight updates. Across Gemma-2-2B/9B and Qwen3-8B, DSPA improves MT-Bench and is competitive on AlpacaEval while preserving multiple-choice accuracy. Under restricted preference data, DSPA remains robust and can rival the two-stage RAHF-SCIT pipeline while requiring up to $4.47\\times$ fewer alignment-stage FLOPs. Finally, we audit the SAE features DSPA modifies, finding that preference directions are dominated by discourse and stylistic signals, and provide theory clarifying the conditional-difference map estimate and when top-$k$ ablation is principled.","url_abs":"https://arxiv.org/abs/2603.21461","url_pdf":"https://arxiv.org/pdf/2603.21461","source":{"archive":null,"snapshot":"2025-07-28","note":"not in the Papers with Code archive (frozen at the snapshot)","row_kind":"graph","title_abstract_authors_date":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)"},"code_links":[],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2603.21461","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2603.21461"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/tatsu-lab/alpaca_eval","reach":null},{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/lm-sys/FastChat","reach":null}],"summary":{"ran_violates":1,"unverified":6},"by_repo_kind":{"found_in_text":{"samples":7,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"0bf414a2b44babe6","entry":"raise_warning_for_incompatible_cpu_offloading_configuration","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/model_adapter.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/model_adapter.py","link_basis":"harvester_set","language":"python","status":"ran_violates","verification_level":1,"contract_check":"VIOLATES","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"0bf414a2b44babe6"}},{"code_sha256_prefix":"80dde39da1ba3708","entry":"compress","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/compression.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/compression.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"80dde39da1ba3708"}},{"code_sha256_prefix":"c5824da57107fe8c","entry":"get_compressed_list","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/compression.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/compression.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"c5824da57107fe8c"}},{"code_sha256_prefix":"0e29f2d151ce2bd7","entry":"get_conv_template","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/conversation.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/conversation.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"0e29f2d151ce2bd7"}},{"code_sha256_prefix":"a42fafd21685fb54","entry":"get_jacobian_trajectory","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/model_cllm.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/model_cllm.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"a42fafd21685fb54"}},{"code_sha256_prefix":"dc21db7cf25bfe72","entry":"process_response","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/model_chatglm.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/model_chatglm.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"dc21db7cf25bfe72"}},{"code_sha256_prefix":"e4f8545686f50507","entry":"recover_message_list","repo":"lm-sys/FastChat","repo_kind":"found_in_text","path":"fastchat/model/model_chatglm.py","file_url":"https://github.com/lm-sys/FastChat/blob/HEAD/fastchat/model/model_chatglm.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e4f8545686f50507"}}]},"arxiv_metadata":{"licence":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)","fields":["title","abstract","authors","date"],"primary_category":"cs.LG","source":"arxiv_2026.jsonl"},"syntology_extracted_results":null}