Foundry

Forge · 0x5D08…0A12

OPENEmbeddingOff-chain manifest

Chemistry-Paper Retrieval Embeddings

A dense embedding model tuned for semantic retrieval over chemistry and materials-science literature, where general-purpose embeddings miss reaction and compound nuance.

Contributions

0

Phase

OPEN

Window

Closed

Creator

0x4f18…CfE8

Contribution window elapsed · view on explorer ↗

About this Forge

What it's training, and why

Researchers searching chemistry corpora are failed by generic embeddings: they collapse distinct reaction mechanisms and confuse compounds with similar names. This Forge fine-tunes an embedding model on contributed query→relevant-passage pairs mined from real literature search sessions, so retrieval ranks the paper a chemist actually wanted. Ownership mints proportionally to each contribution's measured lift in retrieval accuracy.

Base model

bge-base-en-v1.5

Task

Embedding

Fine-tune

LORA

Eval metric

ACCURACY

Holdout size

5,000

Who should contribute

This Forge needs

  • →Chemists and librarians contributing labelled retrieval pairs
  • →Research groups donating anonymised search logs
  • →Compute providers for contrastive fine-tuning
  • →Backers funding annotation of hard negatives

Participate

Where this Forge goes next

Contribution window elapsed
  1. 01

    Contributions are locked

    The window for this Forge has closed — new data/compute/capital can't change ownership of this Ingot.

  2. 02

    Watch attribution settle

    Follow the live attribution panel below as the TEE-attested score vector resolves each contributor's share.

  3. 03

    Use the resulting Ingot

    Once LIVE, run the model via Foundry's OpenAI-compatible proxy — revenue from that usage routes back to contributors.

  4. 04

    Fork it in a new Forge

    Start a downstream Forge from this Ingot. Lineage and attribution carry forward automatically.

Start a new Forge →Read the contributor guide

What you earn

Contribution becomes ownership — provably

Proportional Ingot ownership

Shares of the Ingot NFT mint to you in proportion to your TEE-measured marginal contribution. Not a flat bounty — equity.

On-chain revenue share

Every paid inference call on the Ingot routes a fee through the RevenueSplitter to holders, automatically and forever.

Permanent attribution

Your contribution and its score are logged on 0G — replayable, portable, and inherited by any downstream fork.

Use cases

What the resulting Ingot unlocks

Better literature search

Drop-in embedding upgrade for RAG over chemistry corpora.

Patent & prior-art discovery

Surface mechanistically related work that keyword search misses.

Lineage forks

A downstream Forge can specialise this Ingot for a single sub-field and inherit credit.

Live attribution

Marginal Δ per contribution

Eval has not yet run. Score deltas materialise after the TEE coordinator submits the attested score vector.

No contributions yet. Be the first.

Data

0

Compute

0

Capital

0

Eval coordinator

0x4f18…CfE8

TEE attestation

idle

encrypted holdout

seal0x3b…14a
baseline—
measured—
attestation—

signed score vector

Holdout is decrypted only inside the enclave. The eval result carries a hardware-signed attestation; submitEvalResult on the Forge contract reverts if the attestation is invalid.

Phase

OPEN

Accepting contributions. Data, compute, and capital can be added until the window closes.