rewirebio.iobenchmarks
Model

Carbon-X

Carbon-X reads raw DNA inside a chat conversation and answers in English, predicting variant pathogenicity, classifying sequences and predicting the next bases.

No reviewed evaluations are linked here in this release. See the sources below.

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

Use this model

How it works, versions and access

Additional methodological details, versions and access terms have not yet been extracted into a structured profile for this model. Consult the linked sources and evaluation records below.

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

0 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-02-a7f03c9e55a5
Property and statementOriginal source and locationReview and provenance

No evidence rows match these filters. Choose another scope or clear the search.

Sources and history

Release 2026-11-02-a7f03c9e55a5 · Record review: source checked

2 source records and release historyDownload this release (gzip)
Technical metadata and extraction receipts

Stable ID: finedna-20261101-model-carbon-x

areas
dna-genomes
contexts
research
method types
foundation_model
reported name
Carbon-X
entity level
checkpoint
version
HuggingFaceBio/Carbon-X
checkpoint revision
17de89165e25c49e7de4f845d0e5d464469de359
licence
Apache-2.0
access
Public weights on Hugging Face; 14.8B parameters in bfloat16, served with vLLM on a single 80 GB GPU.
training data
FineDNA (finedna-20261101-data-finedna) for all three stages: 302B tokens of mid-training from Qwen3-14B, then SFT (HuggingFaceBio/Carbon-X-SFT) and GRPO reinforcement learning with verifiable rewards. The card states that all training data was decontaminated against its five evaluation benchmarks by whole-span DNA overlap, text n-gram overlap and sequence homology search.
provenance note
Decoder-only Transformer extended from Qwen3-14B with 5,515 DNA tokens (non-overlapping 6-mers); 40,960-token context. Built by Hugging Face, Zhongguancun Academy and the Telethon Institute of Genetics and Medicine (University of Naples Federico II). Not clinically validated; the card reports weak splice-site (29.8%) and histone-mark (34.6%) accuracy on GenomeQA.
source locator
finedna-20261101-source-carbon-x-card: Model Summary, Key features, Training, Limitations
review
method: source-hash-verification; automated-source-review; ai-assisted-source-review; reviewer: claude; reviewer note: Claude agent; no human review claimed. Extractor and reviewer were the same agent in the same session.; date: 2026-11-01; method note: Both cards re-read at their pinned commits with SHA-256 checked; build.py asserts each quoted count and statement against the card text.; note: Descriptive records only; no result. Verdict in docs/reviews/2026-11-01-finedna.md.
Related records

    Suggest a correction