Upload scripts/train-sakthai-1.5b-v2.py with huggingface_hub

exp: SAK_DUP_CAP + fixed NanGuard (no-labels reforward); sync-tested

NanGuard: zero grads BEFORE diagnostics; dump wrapped in try/except (first run confirmed rows have 6-77 unmasked labels at the poison step)

NanGuard: dump + skip poisoned optimizer steps (no seed is safe; forensics in commit history)

nanhunt: abort at first non-finite gradient, dump the triggering batch (v3 disproved the seed theory)

exp: decouple trainer seed from data seed; trainer seed back to 42 (the NaN cause, not LR)

exp: SAK_RANK/SAK_DROPOUT/SAK_MAXLEN controls to attack argument accuracy

exp: SAK_LR override; 4e-4 diverges to NaN at 6 epochs

exp script: bench-v2 exclusions, SAK_SEED/SAK_EPOCHS/SAK_TAG controls

Add 0.5B experiment script (masking / full-FT modes); eval moves to the bench repo runner

H200: micro-batch 32 x accum 2 (batch 64 OOMs without checkpointing)

Fix validator bench path

Tune 0.5b v7 job for H200: batch 64, lr 4e-4, 3 epochs, no grad checkpointing

Add job-0.5b-v7.py (v7 + bench-v1, trackio crash fixed) and its local validator

0.5B job: 2 epochs, no grad-checkpoint (faster), max_len 1536, checkpoint-to-hub safety

0.5B job: defensive renderer (null messages) + empty-row filter; validated on all 2116 rows

Add HF Jobs 0.5B train+eval script (manual ChatML render, validated)

List new 0.5B train + BFCL eval scripts in Contents

Add BFCL-style eval on the held-out combined-v6 test split

Add improved 0.5B fine-tune script (MLP LoRA targets, rsLoRA, completion-only loss, v6+irrelevance)

Lean rewrite: dynamic download badge, single family table, fix counts/labels, story→profile

Upload README.md with huggingface_hub

Enrich README: current download counts, Rising Stars section, sibling dataset & Spaces sections, ecosystem count fix

Fix static downloads badge → dynamic JSON endpoint; expand related models to full family table

SakSit: professional level-up — House of Sak standards

SakSit: level up to professional standard

SakSit: professional model card — full spec, eval, and House of Sak branding

SakSit: add model card with House of Sak context

Add post-training merge + benchmark script

Add 7B training notebook with validation + benchmark

Endpoint deployment script for merged model

SakThai Engine - Qwen2.5-1.5B LoRA fine-tune notebook

initial commit