Carbon-8B

Model overview

Carbon-8B is the 8B-parameter sibling of Carbon-3B. It is intended for users who can afford additional inference cost in exchange for stronger downstream performance. For the full design rationale, tokenizer specification, evaluation protocol, and usage details, please refer to the **Carbon-3B model card** and the Carbon technical report — this card focuses only on what is specific to Carbon-8B. The `fns` branch loads custom modeling code for Factorized Nucleotide Supervision (FNS). Carbon still uses its efficient 6-mer tokenizer, but during generation each selected 6-mer is assembled from six per-position nucleotide distributions, giving base-pair-level control over decoded DNA. Use this branch when you need exact base-pair counts, per-position masks, or temperature/top-p behavior applied at the nucleotide level rather than over the 4,096-way 6-mer distribution: The same per-base marginals are exposed through `score_sequence()`, which returns the probability assigned to the observed base at each position. Taking the mean log probability gives a base-pair-level sequence score, where higher values indicate higher model likelihood:

Key features

特点摘要待依据来源补齐，不凭模型名称推断能力。

Limitations & considerations

Sources are insufficient to establish specific limitations. Verify task performance, latency, data policies, licensing and version lifecycle.

Source prices & cost analysis

Prices vary by channel, region and configuration. Channel quotes are listed independently and are not treated as a single official provider price.

Public benchmarks & run configuration

暂无已采集的可核验公开评分。模型未上榜不代表能力较低。

Frequently asked questions

Sources & traceability

许可证：apache-2.0。模型介绍与第三方评测均以原来源的具体版本和方法为准。