Commit History

21 days ago

  1. Visualize: selecting a stacked bar keeps its segment colours (only single dots turn pink) Running

    Verified signature
    dad6bef
  2. Visualize: box plot labels get their own band in narrow charts, so dots no longer cover them; label halo

    Verified signature
    4db2af3
  3. Visualize: starter questions carry their plans (no planning call), hedged model calls so one slow provider can't stall a chart, sturdier label discovery, spread samples, whole-number histograms

    Verified signature
    859a42c
  4. Visualize: ask a dataset a question, get an exact chart with the rows behind every mark (small datasets first: instant glance charts, batched model labels, drill-down, scoped follow-ups)

    Verified signature
    917845f

23 days ago

  1. 0.2.2: route through the authenticated optillm proxy (fast tier via DI_MIN_MAX_TOKENS), prompt fixes (plain prose, no invented specifics, no callbacks/placeholders), two-order accuracy-first preference judge, accurate dataset-card provenance

    Verified signature
    bf2993f
  2. 0.2.1: give reasoning models (gpt-oss) token headroom + low reasoning effort, so grading, preference ranking and ping aren't cut off; wheel rebuilt

    Verified signature
    7e4e57b
  3. README: add 'How it compares' (Invent a Dataset, Curation Studio, DataInventor)

    Verified signature
    9e652c4
  4. Default engine: openai/gpt-oss-20b via DeepInfra (more concrete answers, ~2x faster than Llama-3.1-8B in a 5-row test)

    Verified signature
    2803801
  5. Default engine: meta-llama/Llama-3.1-8B-Instruct via DeepInfra on HF Inference Providers (was Qwen3-4B on nscale)

    Verified signature
    13bb7cb

25 days ago

  1. Drop the recreation footer; DataInventor is more than the invent workflow now

    Verified signature
    c77f5f7
  2. Link the Sutra article

    Verified signature
    e964368
  3. Serve the page uncached with versioned assets so embeds pick up new builds

    Verified signature
    545ad61
  4. Host the runner wheel; install from it

    Verified signature
    2bc5774
  5. Curate a corpus: clean, calibrate, create, compose on a sample, plus a recipe and runner for your own machines

    Verified signature
    4f740eb

  1. default to Qwen3-4B-Instruct-2507 on nscale; thinking-off flag now opt-in

    Verified signature
    721a7d3
  2. Add sign-in with Hugging Face and publish-to-my-account with a provenance card

    Verified signature
    4d9dbcd
  3. Fix the endpoint on the host: remove key, URL and model fields; add a Review step; document DI_* settings

    Verified signature
    33e08f1
  4. Remove the hosted-key note from the footer

    Verified signature
    f92de84
  5. Cap hosted runs at 200 rows; offer 25, 50, 100, 200

    Verified signature
    b605207
  6. DataInventor: open-source Invent a Dataset recreation

    Verified signature
    09e451a

  1. Upload Dockerfile with huggingface_hub

    b243c0b
  2. Upload README.md with huggingface_hub

    d5ee673
  3. initial commit

    b6ebb0a