-
feat: update RTX 5090 runner and config
-
refactor: consolidate pure 32L Dense architecture, purge MoE files, audit hardening
-
Update docs/KAGGLE_RUN.md: log_every: 10 and real-time step streaming
-
Audit fix: Update docs/PROGRESS.md for ViuMini-Dense-360M
-
Audit fix: Update docs/KAGGLE_RUN.md for ViuMini-Dense-360M
-
Update docs/HF_PUSH_GUIDE.md for ViuMini-Dense-360M
-
Update docs/PROGRESS.md for ViuMini-Dense-360M
-
Update docs/KAGGLE_RUN.md for ViuMini-Dense-360M
-
Sync docs/PROGRESS.md with massive Mythology and Bollywood Cinema corpora expansion
-
Sync docs/PROGRESS.md with massive Mythology and Bollywood Cinema corpora expansion
-
Sync docs/PROGRESS.md with massive Mythology and Bollywood Cinema corpora expansion
-
Sync docs/PROGRESS.md with massive Mythology and Bollywood Cinema corpora expansion
-
Sync docs/PROGRESS.md with massive Mythology and Bollywood Cinema corpora expansion
-
Sync docs/PROGRESS.md with adult erotica stories priority integration
-
Sync docs/PROGRESS.md with adult erotica stories priority integration
-
feat: step-1 priority queue integration for real Indian history & cricket shards
-
feat: 100% zero-exclusion streaming architecture (2,081 shards) and updated docs
-
feat: 100% zero-exclusion streaming architecture (2,081 shards) and updated docs
-
feat: 100% zero-exclusion streaming architecture (2,081 shards) and updated docs
-
Lock Golden Development Protocol in PROGRESS.md
-
Document Step 1 completion in PROGRESS.md
-
Document RTX 5090 Step 1120 milestone and Viu-1.5B-MoE 42-layer architecture in PROGRESS.md
-
docs: record RTX 5090 live pretraining milestone
-
docs: record tokenizer LFS fix
-
docs: record hardware pricing ROI and rental notebook runners
-
feat: add RTX 5090 & 4090 configs (docs/PROGRESS.md)
-
feat: add RTX 5090 & 4090 configs (docs/RTX_CLOUD_RUN.md)
-
docs: update PROGRESS.md with smart resume and 1k step interval
-
docs: update PROGRESS.md with .gitignore checkpoint fix
-
docs: add hf_split bug resolution and DDP warning explanation
-
docs: update PROGRESS.md with async checkpoints and HF auth warning fix
-
feat: non-blocking asynchronous checkpoint upload and local disk pruning
-
Sync PROGRESS.md: Scale to Global Batch Size 128
-
Sync PROGRESS.md: Unblock toxic and uncensored data (hf_safety: full)
-
Sync PROGRESS.md: Native PyArrow streaming engine
-
Sync PROGRESS.md: Streamlined industry logging update
-
Sync PROGRESS.md: Permanent CastError resolution + direct parquet loader + step 1-20 logging
-
Sync PROGRESS.md: CastError permanent resolution + step <= 10 logging
-
Sync PROGRESS.md: Log full 105B data steps, 10k ckpt, no_sync DDP speedup, and live expert logging
-
Sync PROGRESS.md: train.py CLI args fix
-
Sync docs/PROGRESS.md: Romance literature and casual chat corpus boost
-
feat: wipe-proof HF persistence (full ckpt + live log + newest-2 cleanup)
-
feat: full-ckpt HF push every 10k steps (resume safety net) + docs
-
feat: hf_round corpus-slice rotation (Option A full-data: file rotate + row skip + per-round seed)
-
docs: DDP verify RUN#3 FULL PASS — ready for 50k pretrain
-
fix: exclude non-5col schema paths from HF stream (wikipedia/grammar/gsm8k CastError on shuffle)
-
docs: update PROGRESS.md with Sanatan Dharma, Epics, Gita, Ramayana, and Reasoning boost
-
fix: probe OOM reserve 4GB + expandable_segments + natural-mode stream shuffle (Hindi-only trap)
-
fix: DDP auto_tune probe-before-wrap + broadcast micro/accum, R0-only logs, empty-shard guard, safety import path
-
v1 tokenizer removal notes + v2 download instruction in train config