Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 5, 2025

• - docs: Update README.

  Show description for d958442
  2U1

  committed
  d958442

  Copy full SHA for d958442

Commits on Feb 3, 2025

• feat: support liger-kernel for Qwen2.5-VL.

  2U1

  committed
  6d962d9

  Copy full SHA for 6d962d9

Commits on Feb 2, 2025

• feat: Update for supporting qwen2.5-VL

  2U1

  committed
  4f1ead0

  Copy full SHA for 4f1ead0

Commits on Jan 24, 2025

• feat: Add DoRA option and update training parameters for improved performance

  2U1

  committed
  8c5e253

  Copy full SHA for 8c5e253

• fix: Fix the padding_side to 'right'. It is more efficient when training the LLM model.

  2U1

  committed
  c0c33ff

  Copy full SHA for c0c33ff

• docs: Proper noting for using lora.

  2U1

  committed
  f23ebf6

  Copy full SHA for f23ebf6

• fix: freeze the merger in finetune_lora_vision script to ensure proper training

  2U1

  committed
  402cda4

  Copy full SHA for 402cda4

• fix: properly freeze the base module for lora.

  2U1

  committed
  f9534d6

  Copy full SHA for f9534d6

Commits on Jan 21, 2025

• docs: update README to clarify mixed-dataset support and mark feature as complete

  2U1

  committed
  3c7c3bf

  Copy full SHA for 3c7c3bf

Commits on Jan 20, 2025

• fix: ensure freeze_vision_tower when using vision_lora.

  2U1

  committed
  e107f53

  Copy full SHA for e107f53

• -fix: Refactor dataset for handling mixed-modality dataset.

  Show description for 3298711
  2U1

  committed
  3298711

  Copy full SHA for 3298711

Commits on Jan 18, 2025

• fix: avoid shape mismatch by genertating dummy at the first if-else.

  2U1

  committed
  04b5b13

  Copy full SHA for 04b5b13

Commits on Jan 17, 2025

• fix: support mixed-modality training by updating forward method and dataset handling

  2U1

  committed
  c4d1e7b

  Copy full SHA for c4d1e7b

Commits on Jan 14, 2025

• - fix: fix scripts to use ms-amp and change zero optimization stage

  Show description for 57798fe
  2U1

  committed
  57798fe

  Copy full SHA for 57798fe

Commits on Jan 13, 2025

• fix: update finetuning scripts to use dynamic batch size and gradient accumulation steps

  2U1

  committed
  52be415

  Copy full SHA for 52be415

• fix: change padding side to 'left' for input_ids and labels in DataCollator and update processor default setting

  2U1

  committed
  d2d6140

  Copy full SHA for d2d6140

Commits on Jan 12, 2025

• fix: update min_pixels and max_pixels to original processor_config.

  2U1

  committed
  694380d

  Copy full SHA for 694380d

Commits on Jan 11, 2025

• style: remove commented alternative model name in merge_lora.sh

  2U1

  committed
  5c845bc

  Copy full SHA for 5c845bc

• docs: add SmolVLM Finetune link to README

  2U1

  committed
  10be3e0

  Copy full SHA for 10be3e0

• fix: remove unused metrics parameter from _save_checkpoint method when upgrading to transformers==4.48.0

  2U1

  committed
  52313d0

  Copy full SHA for 52313d0

Commits on Jan 10, 2025

• feat:

  Show description for ab18004
  2U1

  committed
  ab18004

  Copy full SHA for ab18004

Commits on Jan 8, 2025

• fix: ensure freezing of LLM when using LoRA

  2U1

  committed
  a9f324d

  Copy full SHA for a9f324d

Commits on Dec 12, 2024

• fix: Remove unnecessary filtering of state_dict in QwenTrainer

  2U1

  committed
  8922643

  Copy full SHA for 8922643

Commits on Dec 5, 2024

• docs: Update README to include Pixtral Finetune link

  2U1

  committed
  ca3fcd9

  Copy full SHA for ca3fcd9

Commits on Dec 4, 2024

• fix: Update liger-kernel and transformers versions in environment.yaml

  2U1

  committed
  669bf0d

  Copy full SHA for 669bf0d

Commits on Dec 3, 2024

• feat: Update configuration files for optimizer and scheduler settings, and upgrade deepspeed version

  2U1

  committed
  1e34033

  Copy full SHA for 1e34033

Commits on Nov 26, 2024

• fix: Update README and data processing to handle mixed datasets with conditional pixel and grid key checks

  2U1

  committed
  1374853

  Copy full SHA for 1374853

Commits on Nov 5, 2024

• fix: Update finetune_8bit.sh to tune merger instead of projector

  2U1

  committed
  c2f1bc8

  Copy full SHA for c2f1bc8

• docs: Update README and environment.yaml for 8-bit training and dependency changes

  2U1

  committed
  ec4b164

  Copy full SHA for ec4b164

• feat: Add 8-bit finetuning script and update dependencies in environment.yaml

  2U1

  committed
  7ee5e0d

  Copy full SHA for 7ee5e0d

Commits on Oct 29, 2024

• feat: Support text only data.

  2U1

  committed
  80b59b8

  Copy full SHA for 80b59b8

Commits on Oct 10, 2024

• fix: update gradio for not save duplicated images loaded from history in conversation.

  2U1

  committed
  0731c55

  Copy full SHA for 0731c55

Commits on Oct 8, 2024

• fix: update for not saving the processor. It won't change the config for it. So no need to save.

  2U1

  committed
  70f30af

  Copy full SHA for 70f30af

Commits on Oct 4, 2024

• fix: Update enviroment setting for loading proper config.

  2U1

  committed
  7d2eee1

  Copy full SHA for 7d2eee1

Commits on Sep 30, 2024

• feat: Add graio demo.

  2U1

  committed
  ba6f4b7

  Copy full SHA for ba6f4b7

Pagination

Previous

Next