-
docs & scripts: add dedicated grammar stream milestone (145.93GB/105.1B tokens)
-
docs & scripts: add full-scale 5-domain milestone (145.93GB/105B tokens)
-
docs & scripts: add dedicated science stream milestone (128.75GB/92B tokens)
-
docs & scripts: add dictionary, expanded stories, and massive uncensored datasets milestone (117GB/83B tokens)
-
feat/docs: sync stories & uncensored scripts & docs for docs/TEST_REPORT.md
-
feat/docs: sync open-source toxicity datasets & docs for docs/TEST_REPORT.md
-
feat/docs: sync dedicated toxicity folder script & docs for docs/TEST_REPORT.md
-
docs: sync 91.81 GB / 70B token milestone for docs/TEST_REPORT.md
-
docs/feat: sync translation & hinglish expansion for docs/TEST_REPORT.md
-
docs/feat: sync deep domains, wikipedia ingestion & docs for docs/TEST_REPORT.md
-
docs: record 50B+ token master target achieved (65 GB dataset)
-
docs: record 45B token milestone (58.8 GB dataset)
-
docs: record 30B+ token milestone achieved (49.5 GB dataset)
-
fix: Hinglish column key mapping and streamline stream execution order
-
feat: upgrade collector to 1GB shards with smart auto-resume
-
Log verified live Kaggle ingestion metrics (272 shards, 2.72M rows) to TEST_REPORT
-
Add Pillar 26 Grammar & Language Mechanics (Hindi Vyakaran, English Grammar, Hinglish Syntax)
-
Add 25 Pillars: Stories, Movie Plots, Songs, Mobile Tech, Jokes & Street Roasts
-
Add Pillar 21 Translation & Global Cuisines expansion to data pipeline
-
Add Step 4 30B-50B+ Kaggle data collection suite & 20-pillar knowledge matrix
-
feat: SDPA FlashAttention, language tags conditioning, probe safety, slim checkpoints, and streaming clean
-
docs: update TEST_REPORT.md and model/NOTES.md with audit log
-
fix: audit fixes for multi-token alignment, weight decay groups, and MoE monitoring
-
step6: training healthy loss 8.4->5.8
-
step6: OOM fix seq512 batch2 + backup DONE
-
step4: full clean + backup report fixed
-
step5 CLOSED: FINAL tokenizer synced + docs
-
step5: FINAL tokenizer PASS (47756, unk0)
-
step4: word-level tally note
-
step4: clean re-run PASS, step5 ready
-
step4: clean PASS+WARNING, hinglish gap
-
step3 CLOSE: lock 0.05 (yaml+json+code)
-
step3: G loss 3.8244 confirmed
-
step3: F+G PASS, 0.10 lock pending
-
step3: fair-50 C wins, lock pending
-
step3: Config C Kaggle PASS, 0.05 lock pending
-
step3: toy Kaggle PASS A+B, lock pending approval
-
policy: all tests on Kaggle only
-
tests: 5/5 code-phase suite PASS + reports
-
docs: TEST_REPORT + PROGRESS Kaggle smoke
