apunkt/doc-to-lora
1
0
Fork 0
mirror of https://github.com/SakanaAI/doc-to-lora.git synced 2026-07-23 17:01:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Commit graph

  • 696b45c51b icml rebuttal iclr-cleanup 51616 2026-06-15 04:31:47 +00:00
  • 4af1fe8638 Refactor tutorial notebook to extract utility functions and rename TrainArgs to Video2LoRAConfig Sarvesh 2026-06-09 20:56:38 +05:30
  • 538ad2e1ee Add base64 video encoding to comparison dashboard for Colab support Sarvesh 2026-06-09 17:35:29 +05:30
  • 2ae46f8908 Optimize GPU memory usage by running inference in inference_mode and clearing caches Sarvesh 2026-06-09 17:20:14 +05:30
  • a6d558b44d Restructure notebook cells to run base model, video2lora, and visual rendering sequentially Sarvesh 2026-06-09 17:12:12 +05:30
  • 09b6068929 Patch lora forward in modulated model before applying lora to layers in QA inference Sarvesh 2026-06-09 17:03:33 +05:30
  • 51ef4f3d8d Use hasattr keys check to support UserDict-based BatchEncoding Sarvesh 2026-06-09 16:57:30 +05:30
  • 4ac36c1210 Support BatchEncoding in apply_chat_template output during QA inference Sarvesh 2026-06-09 16:48:47 +05:30
  • 073d909468 Reset model hooks prior to base feature extraction and base model inference passes Sarvesh 2026-06-09 16:44:04 +05:30
  • 70e60aa6aa Inject num2words function instead of module into transformers processing_smolvlm Sarvesh 2026-06-09 16:28:32 +05:30
  • e845504d7f Upgrade torchao to 0.16.0 or higher in demo notebook Sarvesh 2026-06-09 16:24:39 +05:30
  • 5a000a7c27 Inject num2words module directly into sys.modules to bypass Hugging Face cache Sarvesh 2026-06-09 16:19:31 +05:30
  • 17f31415d4 Add dynamic num2words cache override to build_stage1_model Sarvesh 2026-06-09 16:15:31 +05:30
  • 4e42b8a1d8 Revert README.md change and write rich, polished markdown explanations inside demo.ipynb Sarvesh 2026-06-09 16:11:09 +05:30
  • 834e979c20 Add num2words dependency to demo notebook and Quick Start section to README Sarvesh 2026-06-09 16:06:49 +05:30
  • 65b6e1cc2c Import hf_hub_download and Path in checkpoint download cell Sarvesh 2026-06-09 16:00:00 +05:30
  • 0b639c84a4 Clean up symlink path bugs and add robust makedirs in demo.ipynb Sarvesh 2026-06-09 15:57:53 +05:30
  • 72e9766f62 Fix Colab clone paths and wrap os.makedirs in try-except in Step 2 Sarvesh 2026-06-09 15:52:46 +05:30
  • 327abd9e38 Configure Step 8 to loop through and display all 6 qualitative examples Sarvesh 2026-06-09 15:51:45 +05:30
  • 2ba3c1bd40 Update demo.ipynb with user structural edits Sarvesh 2026-06-09 15:50:14 +05:30
  • efd61538aa Support cloning website repo to load qualitative videos locally in Colab Sarvesh 2026-06-09 15:48:49 +05:30
  • 482197ab9c Support all 6 qualitative website examples inside demo.ipynb with auto video downloader and HTML5 video player Sarvesh 2026-06-09 15:47:52 +05:30
  • fffc05f4e2 Add run_all_qualitative.py script to run inference on all website qualitative examples Sarvesh 2026-06-09 15:45:58 +05:30
  • 972f44c2fc Auto download sample video if not found locally in demo.ipynb Sarvesh 2026-06-09 15:44:41 +05:30
  • 03c4c96dd9 Support Google Colab auto cloning and setup in demo.ipynb Sarvesh 2026-06-09 15:38:46 +05:30
  • 7fa77b1759 Add self-contained Video2LoRA zero-visual-token inference demo notebook Sarvesh 2026-06-09 15:37:50 +05:30
  • d5648e1b15 Update Sarvesh Baskar personal URL in README.md Sarvesh 2026-06-04 15:36:58 +05:30
  • 3d217fc842 Rename SVG diagram to video2lora-diagram-white.svg to clear GitHub Camo cache Sarvesh 2026-06-04 15:33:52 +05:30
  • a6fb1fb3a6 Bypass GitHub Camo cache for SVG diagram Sarvesh 2026-06-04 15:32:08 +05:30
  • 189a53978d Add white background to the SVG animation diagram Sarvesh 2026-06-04 15:29:25 +05:30
  • eefcca0745 Update README description to official implementation Sarvesh 2026-06-04 15:27:55 +05:30
  • f455cdf2b2 Update README.md with centered header, badges, and quick links Sarvesh 2026-06-04 15:27:13 +05:30
  • 7428582dea Prepare public release Manan Suri 2026-06-03 01:27:45 +00:00
  • baa85db4d5
         Change citation format for Doc-to-LoRA main Rujikorn Charakorn 2026-05-25 16:12:27 +09:00
  • 2e95f3e501
         Add MIT License to the project Rujikorn Charakorn 2026-03-02 13:27:33 +09:00
  • 1abe8ae16d Doc-to-LoRA release 51616 2026-02-27 03:47:04 +00:00
  • 22267c7666 Update README 51616 2026-02-13 08:55:01 +00:00
  • 32b5d2a197 video update 51616 2026-02-13 08:49:48 +00:00
  • 8beaa905e7 Update README demo video section for GitHub rendering 51616 2026-02-13 08:43:49 +00:00
  • ec16f545c1 add bib gif video (waiting for arxiv id) 51616 2026-02-13 08:30:08 +00:00
  • 8093817c34 icml cleaned 51616 2026-02-13 06:02:29 +00:00
  • c4c1ac6eaa cd minibatch working 51616 2025-11-25 04:58:12 +00:00
  • 2e88a18bd2 make ctx encoder vlm compatible 51616 2025-11-10 16:54:55 +09:00
  • 6f857a2dca add generative adapter 51616 2025-11-10 16:54:45 +09:00
  • d6c7e3ab03 fix tracker + recall/precision logging + add generative adapter 51616 2025-11-10 16:54:25 +09:00
  • 2ed418bc22 vlm ctx-encoder + slrum scripts + qwen self-gen 51616 2025-10-27 06:20:26 +00:00
  • 178822b7d3 mistral config + script 51616 2025-10-13 07:45:47 +00:00
  • cddb1238ff self-gen data viewer 51616 2025-10-10 08:09:46 +00:00
  • fc086356b4 add qwen3-4b and mistral-7b (niah working) 51616 2025-10-10 06:31:20 +00:00
  • a074fd3305 self-gen data misaligned fixed + niah reproducible + vllm-tokenizers version combo 51616 2025-10-08 05:17:50 +00:00
  • 39ed1b3d75 basic gradio app 51616 2025-10-02 06:35:33 +00:00
  • 3b798c670c python api + remove generate_multi_lora + rename config path + test eval scripts 51616 2025-10-01 13:11:54 +00:00
  • 530fcd46cc permission + remove copilot intx + remove monkey patch 51616 2025-09-30 15:18:09 +00:00
  • b6679ba755 iclr cleanup 51616 2025-09-30 14:53:21 +00:00
  • 8745db6e11 iclr 2026 (tmp 2) 51616 2025-09-28 15:46:47 +00:00
  • ad18d38374 iclr 2026 (3) 51616 2025-09-28 15:46:09 +00:00
  • 4b1e00fcce Merge branch 'main' of https://github.com/SakanaAI/ctx-to-lora 51616 2025-09-29 00:40:54 +09:00
  • 4f9ba49c04 iclr 2026 (tmp) 51616 2025-09-29 00:40:51 +09:00
  • c21ba2b481 iclr 2026 51616 2025-09-29 00:39:17 +09:00
  • 67e73d0fc7 allow use_bias=False 51616 2025-09-19 06:52:18 +00:00
  • 76bf6981ba scaling gen lora + q gen round for cd + gsm8k 51616 2025-09-18 16:02:22 +09:00
  • 2af651fa41 gsm8k 51616 2025-09-18 16:00:34 +09:00
  • 5d348a4004 truncated eval 51616 2025-09-18 16:00:26 +09:00
  • 8a6c2cfd42 add t2l 51616 2025-09-15 04:04:32 +09:00
  • 3105fac6dc remove num_gen_q 51616 2025-09-15 01:37:28 +09:00
  • b77aa4a9f6 recursive q gen for cd baseline 51616 2025-09-15 01:29:35 +09:00
  • f0721b82a1 old data mix + remove double bos 51616 2025-09-14 16:37:52 +09:00
  • 86ba2bc4a5 self-gen data import 51616 2025-09-13 23:52:24 +09:00
  • 0e46a7c568 add llmlingua + different q model for cd 51616 2025-09-13 23:52:07 +09:00
  • 0e05644d6a iterative mode for hypernet 51616 2025-09-12 19:05:05 +09:00
  • 7d95db6672 context distillation eval + timer 51616 2025-09-12 16:34:03 +09:00
  • f2efe22cba Merge branch 'main' of https://github.com/SakanaAI/ctx-to-lora 51616 2025-09-07 17:16:40 +09:00
  • 650395874f toy exp w/ self-gen + use double bos for eval (following vllm 0.8.4) 51616 2025-09-07 17:16:20 +09:00
  • 0c1a8bb3e8 update data gen/def 51616 2025-09-06 08:43:36 +00:00
  • a4f4dfd522 magic num eval upto 128k 51616 2025-09-06 02:21:15 +09:00
  • cfe737d106 Merge branch 'main' of https://github.com/SakanaAI/ctx-to-lora 51616 2025-09-05 17:53:42 +09:00
  • 23e502aa65 cleaner merger + use_per_ctx_avg_loss=True as default 51616 2025-09-05 17:52:03 +09:00
  • f02ae874f3 sparse kl 51616 2025-09-05 17:51:37 +09:00
  • e2e7f45472 per-ctx-avg loss for ce 51616 2025-09-05 16:28:30 +09:00
  • 237b6dc6fa Merge branch 'main' of https://github.com/SakanaAI/ctx-to-lora into main 51616 2025-09-04 13:54:54 +00:00
  • 1201fcf7bc watcher save state 51616 2025-09-04 13:54:49 +00:00
  • a17d2532c2 fix logging + l1norm div by num ctx (work correctly w/ grad accum) 51616 2025-09-04 21:18:47 +09:00
  • 41a1cfb90a AB scaler + use bias for generation 51616 2025-09-04 21:15:42 +09:00
  • 3364a25591 gcp watcher w/ time filter 51616 2025-09-03 08:51:16 +00:00
  • 563ecf4273 magic alpha! 51616 2025-09-03 17:43:44 +09:00
  • de057cdc30 emb rescale + alpha 1 51616 2025-09-03 13:14:17 +09:00
  • 3505b006ad toy ctx magic num eval upto 32k + remove configs 51616 2025-09-01 05:57:06 +00:00
  • a269aff51d fix empty q_list 51616 2025-08-29 02:36:00 +00:00
  • ee9039295e
         LoRA separate bias param + merger + toy ds (#15) Rujikorn Charakorn 2025-08-29 11:55:24 +09:00
  • 9d82014dc2 q gen can take ds_names + self_gen summ/struct/cot prob + ctx qa v3 data 51616 2025-08-28 17:42:43 +00:00
  • 2e2f0ff473 add tatqa 51616 2025-08-28 09:33:34 +00:00
  • e1a3926e25
         Improved training data generation (#14) ShinnosukeUesakaSakana 2025-08-28 21:03:53 +09:00
  • 1193d10fb4 chat_to_str for chat-history related dataset 51616 2025-08-28 05:17:31 +00:00
  • 7ac74e6d58
         SFT data (#13) Rujikorn Charakorn 2025-08-26 18:14:40 +09:00
  • c30ae19a02 more robust watcher 51616 2025-08-18 09:05:05 +00:00
  • c5e9bc769d
         toy ctx nums and multi-lora training (#11) Rujikorn Charakorn 2025-08-18 18:38:07 +09:00
  • dfee685678 fix web demo with new multi-lora interface 51616 2025-08-13 06:33:34 +00:00
  • 3ae4cb57ce
         use empirical average as offset for sum aggregation (#10) Rujikorn Charakorn 2025-08-13 15:25:50 +09:00
  • 996b89b5a0 chunked eval + backward compat model load 51616 2025-08-12 07:58:36 +00:00
  • 654ddc5b23 only apply split_too_long_qas when ctx_ids is used 51616 2025-08-12 07:58:18 +00:00