docs: add deprecation banner (superseded by -v2/plus-1.5b)

fix: restore valid tokenizer_config.json from base Qwen (broken extra_special_tokens/added_tokens_decoder made model unloadable)

docs: improve model card — YAML inference hints, family badges, usage notes, benchmark honesty

docs: add deployment/llama.cpp/ollama guidance and hardware notes for local inference

Add eval result snapshot via metadata cron on 20260801T055008Z

docs: add How to Use, Benchmarks, and Citation sections

cron: append metadata eval snapshot

health check: sakthai-context-1.5b-tools

docs: timestamp hygiene refresh sakthai-context-1.5b-tools

eval-updater: add metadata eval 2026-07-31T18:55:03Z

Update family count: 20 public models -> 22 public models

Upload .eval_results/benchmark-20260731_104526.yaml with huggingface_hub

Fix stale/incorrect eval results with real measured numbers

eval results on bench-v2 (scorer multiset-selection-v2)

Audit fix: family collection count 20 public models / 15 datasets / 4 Spaces (39 items, bench-v3 joined)

docs: refresh collection/family counts to live state (19 public models, 13 datasets, 4 Spaces)

train.py: DATASET sakthai-combined-v4 (404, never published) -> sakthai-combined-v7 so load_dataset reproduces

Fix broken cross-link: train.py pinned never-published sakthai-combined-v4 (404) -> combined-v7; README provenance note + limitations updated

Improve model card: download badge, model-index (honest unverified bench-v2), architecture table, tool-calling format, training details from train.py, serverless LoRA warning, full 19-model family table, limitations

cron: add eval result for sakthai-context-1.5b-tools (metadata health check, run 5)

Upload README.md with huggingface_hub

Fix cross-link: replace dead sakthai-vision-demo Space (404) with sakthai-web-agent

Upload .eval_results/sakthai-bench-v2.yaml with huggingface_hub

Fix model count 12->14 to match current live count

Add cross-link to context-1.5b-tools-v2, update model count 11→12

fix: add House of Sak narrative, remove broken v7 model link

fix: model count 12→11 (actual from HF API)

fix: update model family count from 13 to 12

cross-link: promote v7 (0 dl) from 1.5b-tools README

fix: correct model count 14→13, remove deprecated English embedding reference

fix: align family count to 14 + add signature tagline

Upload tokenizer

Upload model

docs: align card to lean standard — live downloads badge, single canonical family table, drop funnel sections and stale counts

docs: align card to lean standard — live downloads badge, single canonical family table, drop funnel sections and stale counts

fix: replace broken shields.io/endpoint badge with working static badge (163 dl)

Remove eval_results derived from small internal spot checks, not benchmarks

Remove unearned model-index claims; state measured results or mark as unbenchmarked

Overhaul README: expanded family table (14 models), dataset table (7), Spaces table (3), growing ecosystem section, updated counts

Upload .eval_results/bfcl.yaml with huggingface_hub

Delete .eval_results/bfcl-irrelevance.yaml with huggingface_hub

Delete .eval_results/bfcl-parallel.yaml with huggingface_hub

Delete .eval_results/bfcl-multiple.yaml with huggingface_hub

Delete .eval_results/bfcl-simple.yaml with huggingface_hub

Upload .eval_results/bfcl.yaml with huggingface_hub

Upload .eval_results/bfcl-irrelevance.yaml with huggingface_hub

Upload .eval_results/bfcl-parallel.yaml with huggingface_hub

Upload .eval_results/bfcl-multiple.yaml with huggingface_hub

Upload .eval_results/bfcl-simple.yaml with huggingface_hub

Lean rewrite: dynamic download badge, single family table, honest evals, story→profile