round-10: physics-factorial + direct-2048 chain (src/pino/train.py)

round-10: physics-factorial + direct-2048 chain (src/pino/embeddings.py)

round-7 follow-through: disjoint_256 arm (src/pino/train.py)

round-7 follow-through: disjoint_256 arm (src/pino/embeddings.py)

round-7: morgan_2048_rp parity arm (src/pino/train.py)

round-7: morgan_2048_rp parity arm (src/pino/embeddings.py)

Factorial channel ablation: --physics-mode (gamma_only/headspace_only) in train.py + evaluator matching + factorial job script

Upload src/pino/train.py with huggingface_hub

Upload src/pino/train.py with huggingface_hub

Upload src/pino/pimt_model.py with huggingface_hub

Upload src/pino/train.py with huggingface_hub

review-response: upload src/pino/upload_data.py

review-response: upload src/pino/train.py

Upload src/pino/train.py with huggingface_hub

feat: pom_alltags structural source — 269-dim OpenPOM input, 575-dim tag objective head

fix: pre-existing test failures — draft_engine None-MW crash + stale gate test

feat: low-evidence gating in substitution_v11 + wisemoor description provenance

feat: wire genuine 256-dim OpenPOM into training (--structural-source)

feat: freeze representation validation gate

feat: add leakage-resistant genre baseline benchmark

Complete catalog recovery and enrichment

fix(v11): preregister mode-aware retrieval evaluation

v11 assessment: retriever top5=0.133, descriptor kept, formula scoped

v11 evaluator: substitution eval set n=30, assessment interface

pimt v9 substantivity gbm train publish path

pimt v8 substantivity gate and pubchem cache

Add Fraterworks ingestion and upload data split support

feat: add WiseMoor v9 provenance dataset

fix: normalize objective targets for pyramid training

feat: prepare v7 HF training and evaluation

v6: ConcentrationAwarePyramidHead + OAV fix + curated descriptors + integration

feat: 3-tier scent pyramid objective head + v4 dataset + v5 training job

Lazy-load embedding engine; force HF_REGISTRY_PATH to /data/registry.db

v4: three-tier descriptor targets (v3), collate padding sentinel, HF job v3 dataset

v4: time-aware objective head, physics-derived v2 targets, Pyrfume vocab

fix(loss): prevent objective-starvation with fixed task weights; target pimt_v3.pt

feat: replace sigmoid gating with stabilized FiLM affine modulation; train pimt_v2.pt

fix(hf_job): writable registry + output dirs for read-only source mounts

feat(naturals,shalimar): hierarchical natural oil tokenization

fix(embeddings): handle registry records with empty openpom/shape embeddings

feat(embeddings): expand input token from 145-D to 151-D with physical-functional payload

perf(embeddings): implement write-through SQLite cache for 3D shape features

perf(train): avoid Hugging Face Dataset round-trip during formula-disjoint split

build: lower Python requirement to >=3.10 for official PyTorch image compatibility

feat(semantics): add literature-based odor descriptor vocabulary builder

fix(train): keep AdaptiveLossBalancer forward signature intact, expose last_weights attribute

feat(train): log AdaptiveLossBalancer weights per batch for telemetry verification

chore(train): point default data path to empirical dataset and confirm balancer/loss setup

feat(data,verifier): add skip_ifra to verifier for synthetic dataset generation

feat(data): add commercial solvent dilution layer and clean natural profiles