Upload V13.1 S2 checkpoint

Upload V13.0 S2 checkpoint

Add V12.4.6 video decode manifest

chore: remove S0 placeholder

upload V10.2 S0 main weights

chore: add S0 checkpoint directory

rename ckpts stock_{16f,32f}_10pct -> baseline_* (naming consistency)

add V10.x cache vjepa21/video_10pct_chunks

add V10.x cache qwen3vit/video_10pct_64f

add V10.x cache qwen3vit/video_10pct_32f

add V10.x cache vjepa21/image_10pct

Cleanup: remove flat 4b_stock/4b_v9_1 (moved into S2) + mistaken flat 32f dirs

Reorg: move 16f stock/v9_1 ckpts into S2 stage-first tree (drop machine name, tag 16f)

Reorg: flatten 32f ckpts to ckpts/{v10_2_32f,stock_32f}

Add 4B 32f inference ckpt under stage-first ckpts/ -> ckpts/S2/4b/qwen3_5_2b/stock_32f_10pct/v0-20260606-044138/checkpoint-785

Add 4B 32f inference ckpt under stage-first ckpts/ -> ckpts/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054/checkpoint-785

remove write-path test marker

write-path test marker

docs: fix eval pointer (GitHub README step 7 -> section 4)

Update model card for data/ + ckpts/ + legacy/ restructure

Remove ckpts placeholder now that real ckpts landed

Add 4B MLLM ckpt: 4b_v9_1 (checkpoint-505, inference weights)

Add 4B MLLM ckpt: 4b_stock (checkpoint-505, inference weights)

Create ckpts/ folder for current trained checkpoints

Restructure repo into data/ + legacy/ (batch 14)

Restructure repo into data/ + legacy/ (batch 13)

Restructure repo into data/ + legacy/ (batch 12)

Restructure repo into data/ + legacy/ (batch 11)

Restructure repo into data/ + legacy/ (batch 10)

Restructure repo into data/ + legacy/ (batch 9)

Restructure repo into data/ + legacy/ (batch 8)

Restructure repo into data/ + legacy/ (batch 7)

Restructure repo into data/ + legacy/ (batch 6)

Restructure repo into data/ + legacy/ (batch 5)

Restructure repo into data/ + legacy/ (batch 4)

Restructure repo into data/ + legacy/ (batch 3)

Restructure repo into data/ + legacy/ (batch 2)

Restructure repo into data/ + legacy/ (batch 1)

Add V9 good_manifest (decode-probed video list) for collaborator repro

Add vmllm_cached cached datasets (6.2G, 693 files)

docs: add README describing checkpoint layout and training code mapping

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub