Lmms-engine

What's Changed
* docs: update training launch commands and add hydra override documentation by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/131
* [feat] NanoVLM Training support by Purshow in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/134
* [feat] Add total FLOPS, training duration & CO2 emission tracking by Jinghao-Guo in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/137
* docs: Update raedme.md and scripts of NanoVLM by Purshow in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/138
* [docs] modify qwen3_vl example config by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/140
* [docs] Update NanoVLM readme by Purshow in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/141
* [WIP] feat: Transformers 5.0 compatibility by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/142
* [feat] support messages column as JSON string in iterable datasets by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/147
* [fix] handle parquet schema mismatch in dataset concatenation by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/146
* [feat] Support EP for transformers >= 5.0 by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/148
* doc:fix typo by Purshow in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/149
* [fix] remove dead code, fix typos, and eliminate unnecessary GPU-CPU syncs by choiszt in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/150
* fix(cicd): auto-select free port and fix qwen3 FSDP2 attention_type compat by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/151
* fix(dataset): join data_folder with video path in Qwen3VL iterable dataset by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/152
* chore: add Dockerfile and .dockerignore by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/153
* fix(qwen3_vl): align video timestamp token placement with transformers by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/154
* fix(merger): handle non-sharded tensors in FSDP2 checkpoint merging by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/155
* fix(merger): copy processor_config.json for custom processors by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/156
* fix(merger): re-tie weights to avoid duplicating tied parameters by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/157
* fix(fsdp2): avoid double-dividing tokens_per_gpu by sp_size by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/158
* fix(qwen3_vl): apply pixel limits via processor size by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/159
* fix(merger): use DTensor placements to detect replicated vs sharded params by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/160
* refactor(packing): extract online packing strategies by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/161
* refactor(packing): streaming BalancedPacking to eliminate worker stalls by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/162
* feat(kernels): pluggable varlen attention backend (FA2 / SDPA) by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/163
* fix(packing): rename next_fit to first_fit by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/164
* feat(datasets): support optional video_start/video_end on video_url by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/165
* fix(bagel): guard flash_attn imports so models init works without it by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/166
* feat(qwen3_5): packed linear-attention forward + fix DynamicCache import by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/167
* feat: ViT frame parallel + Qwen3.5 VL forward by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/168
* fix(qwen3_vl): patch vision rope with liger + gate text_model.norm rms_norm by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/169
* [feat] add LLaVA-OneVision2 (8B-Instruct) training support by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/170
* feat(qwen3_5_moe): modeling + EP support + multi-axis DTensor merger by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/171
* fix(iterable-dataset): implement stateful protocol for resume by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/172
* feat(qwen3_5): sequence parallel + ViT frame parallel integration by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/173
* feat(profiler): CUDA memory snapshot with auto-dump on OOM by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/174
* fix(qwen3_5): shard ViT frames across CP ranks by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/175
* fix(profiler): dump memory snapshot on trainer OOM by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/176
* feat: add Qwen3-VL ViT frame parallel SP support by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/177
* feat: add Qwen3-VL tensor parallel by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/178
* feat: add HSDP (Hybrid Sharded Data Parallel) support by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/180
* feat(profiler): add lightweight CUDA event timing by choiszt in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/179
* perf(fsdp2): reduce per-step host/comm overhead by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/181
* docs: fix repo URL in README (LMMs-Lab -> EvolvingLMMs-Lab) by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/182
* perf(rope): vectorize qwen3_vl_get_rope_index, drop per-token sync by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/184
* feat(datasets): lmms_video_utils video backend for LLaVA-OneVision-2 codec stream by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/183
* fix(fsdp2): broadcast non-persistent buffers under enable_parallel by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/185
* feat: add Aero Realtime training support by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/187

New Contributors
* Purshow made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/134
* choiszt made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/150

**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine/compare/v0.1.3...v0.1.4

What's Changed
* doc: fix github link by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/65
* fix: Fix llava ov batched image padding issue by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/72
* [test] add test for qwen2.5omni, fix qwen2.5omni example by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/71
* [feat] support BAGEL training with Liger Kernel by pufanyi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/74
* [docs] Fix BAGEL model packing status in README by pufanyi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/79
* Enhance MFU reference document introduction by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/82
* add linux uv sync script with automatic platform detection by oneScotch in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/81
* [feat] Qwen3 MoE EP Support by Jinghao-Guo in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/75
* [docs] Add Docker usage instructions to README by pangyyyyy in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/85
* add llada and dream arch examples for dllm training by JinjieNi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/84
* [feat] Qwen 3 Omni MOE with EP support by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/88
* [docs] correct doc: diffusion language model by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/89
* Update README.md by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/90
* [fix]: Remove rank == 0 in all makedirs (93) by VietCT04 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/94
* [feat] Qwen 3 VL MOE with EP support by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/92
* [feat] Qwen3 Training support by yiyexy in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/95
* [feat] SP loss better alignment and patch qwen3 vl conv implementation to linear by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/96
* [fix] Handle router logits in Qwen 3 moe and Qwen 3 omni moe for aux loss by ngquangtrung57 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/98
* [feat] LLaVA-Video Training support by nssmd in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/97
* [feat] Gradient accumulation by pufanyi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/103
* Remove linear patch for conv3d for now for precision issue by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/105
* [feat] Allow bagel to output logits and logprobs for sde, fix collator padding for padded images by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/109
* [fix] Update Hydra command for multi-node training by pufanyi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/108
* [fix] Fix some training mismatch in qwen3 vl and rfc parallel logic by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/106
* Add projects using LMMs-Engine to README by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/111
* Fix badge formatting for LongVT project link by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/112
* LLaVAOneVision1_5 Support by Jinghao-Guo in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/101
* [feat] Add map style dataset for qwen3 vl by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/115
* [fix] Align better bagel original eval with option to align with flow-grpo sde settings by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/117
* Update section title and project descriptions in README by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/118
* [feat] enable freeze submodules by gathierry in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/119
* [feat] Better imports utils for lmms-engine by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/122
* [feat] add EMA (Exponential Moving Average) support for FSDP2 training by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/120
* [fix] relax overwrite_config typing to support non-string config overrides by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/124
* Add Bagel Trainer and fix config, bagel data processor by KemingWu in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/126
* [fix] Applied different rnd seed in bagel so that the noise would be sample… by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/129
* [fix]: use valid labels for SP loss normalization by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/130

New Contributors
* oneScotch made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/81
* pangyyyyy made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/85
* JinjieNi made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/84
* KemingWu made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/89
* VietCT04 made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/94
* yiyexy made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/95
* nssmd made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/97
* gathierry made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/119

**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine/compare/v0.1.2...v0.1.3

* fix: Fix launch from cli using config examples by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/54
* feat: Support Qwen2.5 Omni Thinker by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/56
* feat: Add llava_ov, bagel and better cicd readme and control by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/57
* docs: Add a auto build docs, may be deprecated by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/58
* feat: Support Qwen3-VL ulysses sequence parallel operation by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/59
* fix: Fix random shuffle seed on same dp rank to prevent sp hang by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/60
* docs: improve documentation accuracy and add Qwen-VL training guide by mwxely in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/62
* Fix/reorg examples by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/61
* Dev/readme by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/63
* docs: Fix some examples error and better documentation on implementing new class by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/64

New Contributors
* mwxely made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/62

**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine/compare/v0.1.1...v0.1.2

What's Changed
* feat: Custom FSDP2 trainer by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/8
* feat: Add Save and Load logic for fsdp2 trainer by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/9
* Dev/bo 0809 by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/10
* feat: Add flash-attn and liger-kernel dependencies by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/11
* feat: Support Qwen2 for remove padding training by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/14
* [feat] enable dllm training by BIGKnight in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/15
* feat: Add cicd by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/16
* feat: LLaVA-Ov ops and liger-kernel rfc by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/17
* rfc: Better base dataset abstract class and flexible args for kwargs by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/20
* test: Multi-gpu cicd test for robustness by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/22
* Dev/wan by BIGKnight in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/23
* feat: Add Qwen2 ulysses sequence parallel by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/24
* rfc: Refactor video loading logic and processor by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/25
* rfc: Train implementation, monkey patch logic by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/28
* Add efficient loss for dllms by yshenaw in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/27
* feat: Add profiler by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/30
* fix: profile error by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/31
* feat: Support stream packing by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/32
* Dev/muon by BIGKnight in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/34
* fix: Force iterable max steps by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/35
* feat: Support bagel training by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/33
* fix: Image tensor size error by pufanyi in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/40

New Contributors
* yshenaw made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine/pull/27

**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine/compare/v0.1.0...v0.1.1

What's Changed
* feat: Custom FSDP2 trainer by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/8
* feat: Add Save and Load logic for fsdp2 trainer by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/9
* Dev/bo 0809 by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/10
* feat: Add flash-attn and liger-kernel dependencies by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/11
* feat: Support Qwen2 for remove padding training by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/14
* [feat] enable dllm training by BIGKnight in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/15
* feat: Add cicd by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/16
* feat: LLaVA-Ov ops and liger-kernel rfc by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/17


**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/compare/v0.1.0...v0.1.0.post1

What's Changed
* [Feat] simplification by Luodian in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/2
* Dev/fla by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/5
* feat: Refactor Kernels structure and init sp logic by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/1
* fix: Fsdp save and merge by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/6
* feat!: Support ulysses sequence parallel for qwen2_5_vl model by kcz358 in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/7

New Contributors
* Luodian made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/2
* kcz358 made their first contribution in https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/pull/5

**Full Changelog**: https://github.com/EvolvingLMMs-Lab/lmms-engine-mini/commits/v0.1.0