{"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/lyra-an-efficient-and-speech-centric","title":"Lyra: An Efficient and Speech-Centric Framework for Omni-Cognition","arxiv_id":"2412.09501","date":"2024-12-12","proceeding":null,"authors":["Zhisheng Zhong","Chengyao Wang","Yuqi Liu","Senqiao Yang","Longxiang Tang","Yuechen Zhang","Jingyao Li","Tianyuan Qu","Yanwei Li","Yukang Chen","Shaozuo Yu","Sitong Wu","Eric Lo","Shu Liu","Jiaya Jia"],"abstract":"As Multi-modal Large Language Models (MLLMs) evolve, expanding beyond single-domain capabilities is essential to meet the demands for more versatile and efficient AI. However, previous omni-models have insufficiently explored speech, neglecting its integration with multi-modality. We introduce Lyra, an efficient MLLM that enhances multimodal abilities, including advanced long-speech comprehension, sound understanding, cross-modality efficiency, and seamless speech interaction. To achieve efficiency and speech-centric capabilities, Lyra employs three strategies: (1) leveraging existing open-source large models and a proposed multi-modality LoRA to reduce training costs and data requirements; (2) using a latent multi-modality regularizer and extractor to strengthen the relationship between speech and other modalities, thereby enhancing model performance; and (3) constructing a high-quality, extensive dataset that includes 1.5M multi-modal (language, vision, audio) data samples and 12K long speech samples, enabling Lyra to handle complex long speech inputs and achieve more robust omni-cognition. Compared to other omni-methods, Lyra achieves state-of-the-art performance on various vision-language, vision-speech, and speech-language benchmarks, while also using fewer computational resources and less training data.","url_abs":"https://arxiv.org/abs/2412.09501v1","url_pdf":"https://arxiv.org/pdf/2412.09501v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"lyra-an-efficient-and-speech-centric","repo_url":"https://github.com/dvlab-research/Lyra","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":null,"task_name":"EgoSchema"},{"task_slug":null,"task_name":"MM-Vet"},{"task_slug":"mme","task_name":"MME"},{"task_slug":null,"task_name":"MVBench"},{"task_slug":null,"task_name":"TextVQA"},{"task_slug":null,"task_name":"Video MME"},{"task_slug":"visual-question-answering-1","task_name":"Visual Question Answering"},{"task_slug":"visual-question-answering","task_name":"Visual Question Answering (VQA)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/visual-question-answering-on-mm-vet","task":"Visual Question Answering","dataset":"MM-Vet","model":"Lyra-Pro","rank_in_archive_order":9,"of":231,"metrics":{"GPT-4 score":"71.4","Params":"74B"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-on-mm-vet","task":"Visual Question Answering","dataset":"MM-Vet","model":"Lyra-Base","rank_in_archive_order":27,"of":231,"metrics":{"GPT-4 score":"63.5","Params":"9B"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-on-mm-vet","task":"Visual Question Answering","dataset":"MM-Vet","model":"Lyra-Mini","rank_in_archive_order":59,"of":231,"metrics":{"GPT-4 score":"51.2","Params":"3B"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-egoschema","task":"Visual Question Answering (VQA)","dataset":"EgoSchema","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"75.8"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-mm-vet","task":"Visual Question Answering (VQA)","dataset":"MM-Vet","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"71.4"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-mme","task":"Visual Question Answering (VQA)","dataset":"MME","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"2485"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-mvbench","task":"Visual Question Answering (VQA)","dataset":"MVBench","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"72.3"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-textvqa","task":"Visual Question Answering (VQA)","dataset":"TextVQA","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"83.5"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-vqa-on-video-mme","task":"Visual Question Answering (VQA)","dataset":"Video MME","model":"Lyra-Pro","rank_in_archive_order":1,"of":1,"metrics":{"Acc":"69.9"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2412.09501","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2412.09501"}},"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":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/dvlab-research/Lyra","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran_fixture":1,"ran_draft_wrong":2,"unverified":16},"by_repo_kind":{"official":{"samples":19,"ran":3,"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":"b0a851cae92754ff","entry":"extract_patches","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/mm_utils.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/mm_utils.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"b0a851cae92754ff"}},{"code_sha256_prefix":"42a46570620cd9fa","entry":"get_chunk","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/eval/model_lyra_image_speech.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/eval/model_lyra_image_speech.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"well_formed","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"42a46570620cd9fa"}},{"code_sha256_prefix":"076c252c52cbb161","entry":"split_list","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/eval/model_lyra_image_speech.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/eval/model_lyra_image_speech.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"076c252c52cbb161"}},{"code_sha256_prefix":"3c8823948724e269","entry":"auto_pad_images","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/mm_utils.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/mm_utils.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":"3c8823948724e269"}},{"code_sha256_prefix":"23e91c5c56d7ff80","entry":"build_unit_tokenizer","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/serve/model_worker.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/serve/model_worker.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":"23e91c5c56d7ff80"}},{"code_sha256_prefix":"34f95bf62dff3ebd","entry":"compute_softdtw","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/soft_dtw_cuda.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/soft_dtw_cuda.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":"34f95bf62dff3ebd"}},{"code_sha256_prefix":"ec64f40bfc12544a","entry":"compute_softdtw_backward","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/soft_dtw_cuda.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/soft_dtw_cuda.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":"ec64f40bfc12544a"}},{"code_sha256_prefix":"68283f6ff784cc32","entry":"forward_attn","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/qwen2vl_top_attn.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/qwen2vl_top_attn.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":"68283f6ff784cc32"}},{"code_sha256_prefix":"aebe8894bde0f55e","entry":"forward_layer","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/qwen2vl_top_attn.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/qwen2vl_top_attn.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":"aebe8894bde0f55e"}},{"code_sha256_prefix":"74e2126c3223cdd8","entry":"forward_model","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/qwen2vl_top_attn.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/qwen2vl_top_attn.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":"74e2126c3223cdd8"}},{"code_sha256_prefix":"f06df1307d1b0d63","entry":"img_to_base64","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/conversation.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/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":"f06df1307d1b0d63"}},{"code_sha256_prefix":"91cfc0ec83571808","entry":"initialize_input_ids_for_generation","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/builder.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/builder.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":"91cfc0ec83571808"}},{"code_sha256_prefix":"34782af36cf765e7","entry":"mov_to_base64","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/conversation.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/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":"34782af36cf765e7"}},{"code_sha256_prefix":"37899f22fb191b37","entry":"pretty_print_semaphore","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/utils.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/utils.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":"37899f22fb191b37"}},{"code_sha256_prefix":"c80358dff96c25ce","entry":"resize_and_center_crop","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/mm_utils.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/mm_utils.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":"c80358dff96c25ce"}},{"code_sha256_prefix":"7719de171ef38b85","entry":"save_result","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/eval/model_lyra_text_speech.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/eval/model_lyra_text_speech.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":"7719de171ef38b85"}},{"code_sha256_prefix":"89a318a54beb022e","entry":"timed_run","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/model/soft_dtw_cuda.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/model/soft_dtw_cuda.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":"89a318a54beb022e"}},{"code_sha256_prefix":"f9939a84b9a65279","entry":"violates_moderation","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/utils.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/utils.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":"f9939a84b9a65279"}},{"code_sha256_prefix":"dc99ac2d1a2f8f6b","entry":"wav_to_base64","repo":"dvlab-research/Lyra","repo_kind":"official","path":"lyra/conversation.py","file_url":"https://github.com/dvlab-research/Lyra/blob/HEAD/lyra/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":"dc99ac2d1a2f8f6b"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}