| { |
| "_generated_at": "2026-10-22T07:35:25.048151", |
| "repo": "BashkirNLPWorld/bashkir-nlp-benchmark", |
| "repo_type": "model", |
| "paper": "BashkirNLP: A Comprehensive Resource Suite and Benchmark for Parameter- Efficient Fine-Tuning of Low-Resource Agglutinative Languages", |
| "uploaded": { |
| "results_files": [ |
| "results/master/MASTER_REPORT_10k.xlsx", |
| "results/master/MASTER_REPORT_10k.json", |
| "results/master/RELEASE_README_10k.md", |
| "results/master/tokenizer_comparison_table_release.csv", |
| "results/master/DEPRECATED_FILES.txt", |
| "results/step9/step9_final_summary_10k.json", |
| "results/step9/tab_rag_mad_10k.tex", |
| "results/step9/rag_detail_10k.csv", |
| "results/step9/mad_detail_10k.csv", |
| "results/step7/human_validation_batch1.json", |
| "results/step7/human_stats_batch1.json", |
| "results/step7/tab_human.tex", |
| "results/step7/tab_human_tests.tex", |
| "results/paper_tables/step3/lm_cross_model.csv", |
| "results/paper_tables/step3/tab_lm_stats.tex", |
| "results/paper_tables/step4/cpt_ppl_both.csv", |
| "results/paper_tables/step4/cls_all.csv", |
| "results/paper_tables/step4/tab_cpt_lm_vs_cls.tex", |
| "results/paper_tables/step4/tab_binary_cross_model.tex", |
| "results/paper_tables/step4/tab_multiclass_cross_model.tex", |
| "results/paper_tables/step5/tokenizer_metrics.csv", |
| "results/paper_tables/step5/tab_morph.tex", |
| "results/paper_tables/step5/tab_fertility_clean.tex", |
| "results/paper_tables/step6/cls_metrics.csv", |
| "results/paper_tables/step6/binary_cross_model.csv", |
| "results/paper_tables/step6/multiclass_cross_model.csv", |
| "results/paper_tables/step6/binary_majority.csv", |
| "results/paper_tables/step6/multiclass_majority.csv", |
| "results/paper_tables/step6/tab_binary_cross_model_stats.tex", |
| "results/paper_tables/step6/tab_multiclass_cross_model_stats.tex", |
| "results/paper_tables/step6/tab_binary_majority.tex", |
| "results/paper_tables/step6/tab_multiclass_majority.tex", |
| "results/step8/holm_bonferroni.csv", |
| "results/step8/variance_decomposition.csv" |
| ], |
| "splits_files": [ |
| "splits/corpus/train.csv", |
| "splits/corpus/validation.csv", |
| "splits/corpus/test.csv", |
| "splits/corpus/metadata.json", |
| "splits/corpus/class_distribution.json", |
| "splits/binary/train.csv", |
| "splits/binary/validation.csv", |
| "splits/binary/test.csv", |
| "splits/binary/metadata.json", |
| "splits/binary/class_distribution.json", |
| "splits/multiclass/train.csv", |
| "splits/multiclass/validation.csv", |
| "splits/multiclass/test.csv", |
| "splits/multiclass/metadata.json", |
| "splits/multiclass/class_distribution.json", |
| "splits/multilabel/train.csv", |
| "splits/multilabel/validation.csv", |
| "splits/multilabel/test.csv", |
| "splits/multilabel/metadata.json" |
| ], |
| "adapters_dirs": [ |
| "adapters/step4-cpt/cpt_mistral-7b_lora_r16", |
| "adapters/step4-cpt/cpt_yandexgpt-5-lite_lora_r16", |
| "adapters/step4-cpt/cpt_phi-3-mini_lora_r16", |
| "adapters/step4-downstream/downstream_binary_mistral-7b_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_binary_mistral-7b_qlora_no_cpt_seed42", |
| "adapters/step4-downstream/downstream_binary_yandexgpt-5-lite_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_binary_yandexgpt-5-lite_qlora_no_cpt_seed42", |
| "adapters/step4-downstream/downstream_binary_phi-3-mini_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_binary_phi-3-mini_qlora_no_cpt_seed42", |
| "adapters/step4-downstream/downstream_multiclass_mistral-7b_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_multiclass_mistral-7b_qlora_no_cpt_seed42", |
| "adapters/step4-downstream/downstream_multiclass_yandexgpt-5-lite_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_multiclass_yandexgpt-5-lite_qlora_no_cpt_seed42", |
| "adapters/step4-downstream/downstream_multiclass_phi-3-mini_cpt_then_qlora_seed42", |
| "adapters/step4-downstream/downstream_multiclass_phi-3-mini_qlora_no_cpt_seed42", |
| "adapters/step6-qlora-cls/cls_binary_yandexgpt-5-lite_seed42", |
| "adapters/step6-qlora-cls/cls_binary_mistral-7b_seed42", |
| "adapters/step6-qlora-cls/cls_binary_phi-3-mini_seed42", |
| "adapters/step6-qlora-cls/cls_multiclass_yandexgpt-5-lite_seed42", |
| "adapters/step6-qlora-cls/cls_multiclass_mistral-7b_seed42", |
| "adapters/step6-qlora-cls/cls_multiclass_phi-3-mini_seed42" |
| ], |
| "encoders_dirs": [] |
| }, |
| "missing": { |
| "reason_notes": { |
| "distilgpt2_gpt2_gpt2medium_full_ft": "Weights not saved — SAVE_FINAL_MODEL=False in paper config.", |
| "step1_step3_classification": "Not in paper — только step4 и step6 CLS.", |
| "deepseek_qwen_phi2": "Not in downstream — исключены в §3.4 / §6.6.", |
| "encoders_full_ft": "step6 не сохраняет encoder weights (только predictions). Reproduction requires SAVE_CLS_MODEL=True.", |
| "paper_tables": "Если отсутствуют — перезапустите соответствующий stepN_results.py (paper_tables сохраняются в LOCAL и стираются при рестарте Colab).", |
| "step3_qlora_lm_adapters": "Если отсутствуют — step3 LM адаптеры не сохранились (возможно, перезапуск). Перезапустите step3 с SAVE_FINAL_MODEL=True." |
| }, |
| "files": [ |
| { |
| "path": "results/paper_tables/step1/lm_aggregated.csv", |
| "tried": [ |
| "step1_full_ft_10k/paper_tables/lm_aggregated.csv" |
| ] |
| }, |
| { |
| "path": "results/paper_tables/step1/cls_aggregated.csv", |
| "tried": [ |
| "step1_full_ft_10k/paper_tables/cls_aggregated.csv" |
| ] |
| }, |
| { |
| "path": "results/paper_tables/step2/icl_aggregated.csv", |
| "tried": [ |
| "step2_icl_10k/paper_tables/icl_aggregated.csv" |
| ] |
| }, |
| { |
| "path": "results/paper_tables/step2/icl_stats_binary.csv", |
| "tried": [ |
| "step2_icl_10k/paper_tables/icl_stats_binary.csv" |
| ] |
| }, |
| { |
| "path": "results/paper_tables/step2/tab_icl_stats.tex", |
| "tried": [ |
| "step2_icl_10k/paper_tables/tab_icl_stats.tex" |
| ] |
| } |
| ], |
| "folders": [ |
| { |
| "path": "adapters/step3-qlora/lm_mistral-7b_qlora_r16_seed42", |
| "tried": [ |
| "step3_peft_10k/adapters/lm_mistral-7b_qlora_r16_seed42" |
| ] |
| }, |
| { |
| "path": "adapters/step3-qlora/lm_yandexgpt-5-lite_qlora_r16_seed42", |
| "tried": [ |
| "step3_peft_10k/adapters/lm_yandexgpt-5-lite_qlora_r16_seed42" |
| ] |
| }, |
| { |
| "path": "adapters/step3-qlora/lm_phi-3-mini_qlora_r16_seed42", |
| "tried": [ |
| "step3_peft_10k/adapters/lm_phi-3-mini_qlora_r16_seed42" |
| ] |
| }, |
| { |
| "path": "encoders/mbert/cls_binary_mbert_seed42", |
| "tried": [ |
| "step6_classification_10k/ckpt/cls_binary_mbert_seed42", |
| "step6_classification_10k/models/cls_binary_mbert_seed42" |
| ] |
| }, |
| { |
| "path": "encoders/mbert/cls_multiclass_mbert_seed42", |
| "tried": [ |
| "step6_classification_10k/ckpt/cls_multiclass_mbert_seed42", |
| "step6_classification_10k/models/cls_multiclass_mbert_seed42" |
| ] |
| }, |
| { |
| "path": "encoders/xlm-r/cls_binary_xlm-r_seed42", |
| "tried": [ |
| "step6_classification_10k/ckpt/cls_binary_xlm-r_seed42", |
| "step6_classification_10k/models/cls_binary_xlm-r_seed42" |
| ] |
| }, |
| { |
| "path": "encoders/xlm-r/cls_multiclass_xlm-r_seed42", |
| "tried": [ |
| "step6_classification_10k/ckpt/cls_multiclass_xlm-r_seed42", |
| "step6_classification_10k/models/cls_multiclass_xlm-r_seed42" |
| ] |
| } |
| ] |
| }, |
| "counts": { |
| "uploaded_results": 34, |
| "uploaded_splits": 19, |
| "uploaded_adapters": 21, |
| "uploaded_encoders": 0, |
| "missing_files": 5, |
| "missing_dirs": 7, |
| "failed_uploads": 0 |
| } |
| } |
