fix(pyproject): add tensorboard dependency for HF job

fix(hf_job): install full dependencies instead of --no-deps

fix(hf_job): download dataset from Hub and point to clean source repo

chore(hf_job): redact token from remote and clean device print

fix(pyproject): pin transformers to stable 4.x line and add HF job artifacts

fix(upload,train): use load_dataset select and fingerprint-based leakage check

feat(upload,train): formula-disjoint split with full corpus recovery

fix(upload,train): enforce strict molecule-disjoint isolation

refactor(train): directive-style molecule-disjoint data ingestion

feat(upload,train): directive-compliant molecule-disjoint split

chore(train): default DataLoader workers to 0 for local check cycles

feat(train,upload): multi-task training with InfoNCE and molecule-disjoint upload

feat(data): generate OAV targets in dataset builder and validate UNIFAC subgroups

refactor(vle): move UNIFAC VLE implementation into pino.thermo module

feat(semantics): compute OAV-weighted descriptor targets from Pyrfume data

feat(pimt_model): expose tokens/states/mask forward and emit genre labels

feat(heads): decompose subjective head and add contrastive projection

feat(data): rebalance text conditioning toward structured Poucher profiles

feat(data): wire structured Poucher profiles into text conditioning pipeline

feat(literature): add structured Poucher odour profile parser

fix(training): update hf_train_job to run from cloned repo, add text conditioning, fix dataset viewer

fix(upload): cast genre to ClassLabel; align PIMT forward with fused labels and publication metrics

feat(pino): publication-grade training telemetry and fused multi-task labels

chore(pino): add run_training entry point for Space compatibility

feat(pino): literature seed integration, stratified dataset split, registry enrichment, and Space sync

feat(literature): add trade-name resolver, Poucher/Arctander parsers, and PyMuPDF extraction scripts

feat(ocr): add Unlimited-OCR Space client and keep registry.db out of git

feat(literature): ingest and seed human-designed fragrance skeletons

fix(generate_dataset): support --total-records flag and default to 5000

feat(scraper): TGSC enrichment with mmHg->Pa conversion and priority registry updates

feat(optimizer): masked MSE, strategy seeds, adversarial IFRA guard; add accords/materials/importer

perf(optimizer): cache PIMT model locally instead of hitting Space every candidate

feat: add Milestone 5 generative inverse-design composer (CMA-ES)

fix: skip non-formulation records in dataset loader; keep feedback separate from dataset JSONL

chore: remove binary registry.db from repo

feat: Milestone 4 clean code snapshot