{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:AWBWG7T2LTPT5NTGNFJKGBBKTG","short_pith_number":"pith:AWBWG7T2","canonical_record":{"source":{"id":"2605.04948","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-06T14:14:35Z","cross_cats_sorted":[],"title_canon_sha256":"35df3c614deb2de9e8d5bcdbb2c94648e245f7d69825301ca81b82508341b446","abstract_canon_sha256":"42bfc4c333f2ae086c17dbfae01adefb3b03da57166a538f8b573bd63b08a958"},"schema_version":"1.0"},"canonical_sha256":"0583637e7a5cdf3eb6666952a3042a99b5f508b05fe35d8ac191b6f1a665396a","source":{"kind":"arxiv","id":"2605.04948","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04948","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04948v2","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04948","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_12","alias_value":"AWBWG7T2LTPT","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_16","alias_value":"AWBWG7T2LTPT5NTG","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_8","alias_value":"AWBWG7T2","created_at":"2026-06-02T03:05:05Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:AWBWG7T2LTPT5NTGNFJKGBBKTG","target":"record","payload":{"canonical_record":{"source":{"id":"2605.04948","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-06T14:14:35Z","cross_cats_sorted":[],"title_canon_sha256":"35df3c614deb2de9e8d5bcdbb2c94648e245f7d69825301ca81b82508341b446","abstract_canon_sha256":"42bfc4c333f2ae086c17dbfae01adefb3b03da57166a538f8b573bd63b08a958"},"schema_version":"1.0"},"canonical_sha256":"0583637e7a5cdf3eb6666952a3042a99b5f508b05fe35d8ac191b6f1a665396a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T03:05:05.623888Z","signature_b64":"JdBtIrlM9QOzKuwHkBGTn6bso8vFOIQVHkJL03XEt4d3SlkfLV58+ZSW24ECSLlf2zeAVtUrTFaF/C1mdrzMAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0583637e7a5cdf3eb6666952a3042a99b5f508b05fe35d8ac191b6f1a665396a","last_reissued_at":"2026-06-02T03:05:05.623385Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T03:05:05.623385Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.04948","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T03:05:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4rxZmtXjUL5Jd2lZy/j0O6n178thglRQ0dTSOyqwZi0PYCOzT+yn0ngnPPDGnCyz4NMZ8gmQNvCKPVmNOK7VAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T06:22:00.205466Z"},"content_sha256":"aee626edc917d10861396dc46478ab2fb5b47b66bd116168ba42eef7fa480496","schema_version":"1.0","event_id":"sha256:aee626edc917d10861396dc46478ab2fb5b47b66bd116168ba42eef7fa480496"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:AWBWG7T2LTPT5NTGNFJKGBBKTG","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Adapting Large Language Models to a Low-Resource Agglutinative Language: A Comparative Study of LoRA and QLoRA for Bashkir","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Mullosharaf K. Arabov, Svetlana S. Khaybullina","submitted_at":"2026-05-06T14:14:35Z","abstract_excerpt":"This paper presents a comparative study of parameter-efficient fine-tuning (PEFT) methods, including LoRA and QLoRA, applied to the task of adapting large language models to the Bashkir language, a low-resource agglutinative language of the Turkic family. Experimental evaluation is conducted on a Bashkir text corpus of 71k documents (46.9M tokens) using models of various architectures: DistilGPT2, GPT-2 (base, medium), Phi-2, Qwen2.5-7B, DeepSeek-7B, and Mistral-7B. To improve the reliability of results, each configuration was trained with three different random seeds.\n  The lowest perplexity "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 71k-document Bashkir corpus is sufficiently representative for both training and reliable test-set evaluation, and that tokenizer compatibility issues do not dominate the observed performance differences across architectures.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"QLoRA on 7B-scale models like Mistral achieves perplexity within 0.45 of full fine-tuning on GPT-2 medium for Bashkir while using over 40 times fewer trainable parameters, though best perplexity does not guarantee coherent monolingual generation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"46943ad90172ea4a60818ff0e8f103566e727a335fa8e6c235f8b28f16873923"},"source":{"id":"2605.04948","kind":"arxiv","version":2},"verdict":{"id":"152a888a-7329-433d-ba19-df3d6e51b91f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T16:18:49.365289Z","strongest_claim":"QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters.","one_line_summary":"QLoRA on 7B-scale models like Mistral achieves perplexity within 0.45 of full fine-tuning on GPT-2 medium for Bashkir while using over 40 times fewer trainable parameters, though best perplexity does not guarantee coherent monolingual generation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 71k-document Bashkir corpus is sufficiently representative for both training and reliable test-set evaluation, and that tokenizer compatibility issues do not dominate the observed performance differences across architectures.","pith_extraction_headline":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.04948/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T10:40:00.944081Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:31:19.942052Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T13:59:48.104064Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"377890cc0a23b33606347a331abcf43fb74b14ec8d7acbc819db724112913c01"},"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":"152a888a-7329-433d-ba19-df3d6e51b91f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T03:05:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"FW7xvFpaAw5o8zNntAUeMOEZeNf9QoPoSL8yMq90usTTXhFNa7YTmhrTao8ybntzHrkgQ4QpgkVFRkQEilFFBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T06:22:00.206266Z"},"content_sha256":"3489beaad576ffc2c40735924fdeb21f80ed972b9695e54f024d5e200c53a520","schema_version":"1.0","event_id":"sha256:3489beaad576ffc2c40735924fdeb21f80ed972b9695e54f024d5e200c53a520"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/bundle.json","state_url":"https://pith.science/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/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":"2026-07-01T06:22:00Z","links":{"resolver":"https://pith.science/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG","bundle":"https://pith.science/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/bundle.json","state":"https://pith.science/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/state.json","well_known_bundle":"https://pith.science/.well-known/pith/AWBWG7T2LTPT5NTGNFJKGBBKTG/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:AWBWG7T2LTPT5NTGNFJKGBBKTG","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":"42bfc4c333f2ae086c17dbfae01adefb3b03da57166a538f8b573bd63b08a958","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-06T14:14:35Z","title_canon_sha256":"35df3c614deb2de9e8d5bcdbb2c94648e245f7d69825301ca81b82508341b446"},"schema_version":"1.0","source":{"id":"2605.04948","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.04948","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"arxiv_version","alias_value":"2605.04948v2","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.04948","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_12","alias_value":"AWBWG7T2LTPT","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_16","alias_value":"AWBWG7T2LTPT5NTG","created_at":"2026-06-02T03:05:05Z"},{"alias_kind":"pith_short_8","alias_value":"AWBWG7T2","created_at":"2026-06-02T03:05:05Z"}],"graph_snapshots":[{"event_id":"sha256:3489beaad576ffc2c40735924fdeb21f80ed972b9695e54f024d5e200c53a520","target":"graph","created_at":"2026-06-02T03:05:05Z","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":"QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the 71k-document Bashkir corpus is sufficiently representative for both training and reliable test-set evaluation, and that tokenizer compatibility issues do not dominate the observed performance differences across architectures."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"QLoRA on 7B-scale models like Mistral achieves perplexity within 0.45 of full fine-tuning on GPT-2 medium for Bashkir while using over 40 times fewer trainable parameters, though best perplexity does not guarantee coherent monolingual generation."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters."}],"snapshot_sha256":"46943ad90172ea4a60818ff0e8f103566e727a335fa8e6c235f8b28f16873923"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T10:40:00.944081Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T21:31:19.942052Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T13:59:48.104064Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.04948/integrity.json","findings":[],"snapshot_sha256":"377890cc0a23b33606347a331abcf43fb74b14ec8d7acbc819db724112913c01","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"This paper presents a comparative study of parameter-efficient fine-tuning (PEFT) methods, including LoRA and QLoRA, applied to the task of adapting large language models to the Bashkir language, a low-resource agglutinative language of the Turkic family. Experimental evaluation is conducted on a Bashkir text corpus of 71k documents (46.9M tokens) using models of various architectures: DistilGPT2, GPT-2 (base, medium), Phi-2, Qwen2.5-7B, DeepSeek-7B, and Mistral-7B. To improve the reliability of results, each configuration was trained with three different random seeds.\n  The lowest perplexity ","authors_text":"Mullosharaf K. Arabov, Svetlana S. Khaybullina","cross_cats":[],"headline":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-06T14:14:35Z","title":"Adapting Large Language Models to a Low-Resource Agglutinative Language: A Comparative Study of LoRA and QLoRA for Bashkir"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.04948","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-08T16:18:49.365289Z","id":"152a888a-7329-433d-ba19-df3d6e51b91f","model_set":{"reader":"grok-4.3"},"one_line_summary":"QLoRA on 7B-scale models like Mistral achieves perplexity within 0.45 of full fine-tuning on GPT-2 medium for Bashkir while using over 40 times fewer trainable parameters, though best perplexity does not guarantee coherent monolingual generation.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"QLoRA on Mistral-7B and Phi-2 reaches near full fine-tuning perplexity for Bashkir with over 40 times fewer trainable parameters.","strongest_claim":"QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters.","weakest_assumption":"That the 71k-document Bashkir corpus is sufficiently representative for both training and reliable test-set evaluation, and that tokenizer compatibility issues do not dominate the observed performance differences across architectures."}},"verdict_id":"152a888a-7329-433d-ba19-df3d6e51b91f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:aee626edc917d10861396dc46478ab2fb5b47b66bd116168ba42eef7fa480496","target":"record","created_at":"2026-06-02T03:05:05Z","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":"42bfc4c333f2ae086c17dbfae01adefb3b03da57166a538f8b573bd63b08a958","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-06T14:14:35Z","title_canon_sha256":"35df3c614deb2de9e8d5bcdbb2c94648e245f7d69825301ca81b82508341b446"},"schema_version":"1.0","source":{"id":"2605.04948","kind":"arxiv","version":2}},"canonical_sha256":"0583637e7a5cdf3eb6666952a3042a99b5f508b05fe35d8ac191b6f1a665396a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0583637e7a5cdf3eb6666952a3042a99b5f508b05fe35d8ac191b6f1a665396a","first_computed_at":"2026-06-02T03:05:05.623385Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-02T03:05:05.623385Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"JdBtIrlM9QOzKuwHkBGTn6bso8vFOIQVHkJL03XEt4d3SlkfLV58+ZSW24ECSLlf2zeAVtUrTFaF/C1mdrzMAQ==","signature_status":"signed_v1","signed_at":"2026-06-02T03:05:05.623888Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.04948","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:aee626edc917d10861396dc46478ab2fb5b47b66bd116168ba42eef7fa480496","sha256:3489beaad576ffc2c40735924fdeb21f80ed972b9695e54f024d5e200c53a520"],"state_sha256":"6dcda977cf7084f0f016e113577332547297a5aa5b78abd1d4fc7e30828c1f7d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"msZhqU5vvq7TGmrmM0ZcJ+8SI51IuBm/NJc7IHm2jepRdW8gpGtNGHjaRQvFRKhVySt53KjEh9H7ArXaiiXoCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-01T06:22:00.209720Z","bundle_sha256":"3f1eacca0a6ab585781d0d5300d8b0439b2f76ed7ed0c9f34ca946426acb802a"}}