Open Models Almanac
Methodspost-trainingmid-training & continual pretraining

specific method · filed under post-training

Software-engineering mid-training

Mid-training on software-engineering materials such as repositories, issue–pull-request pairs, build or CI logs, and agent trajectories.

The sources on this page name it without explaining how it works. The line above is the extraction's one-line gloss. See mid-training & continual pretraining for the methods filed beside it.

Also called Agentic software-engineering mid-training.

sources
4
models
2
lab adopt it
1
strongest
used

Documented in

Evidence

5 quotes from 4 sources, one card per model, each opening with its strongest.

How sources treat it
used 5

A 65.4-billion-token stage at 256K context uses repositories, issue–pull-request pairs, commit histories, build and CI logs, and agent trajectories.

usedtraining objectivein Nemotron 3.5 SuperNVIDIA
4 more from this model

The mid-training mixture combines whole repositories, issue–pull-request pairs, commit histories, build and CI logs, and agent trajectories.

useddata curationin Nemotron 3.5 SuperNVIDIA

We mid-trained the long-context checkpoint for 65.4B tokens at a sequence length of 256K.

usedtraining objectivein Nemotron 3.5 SuperNVIDIA

67B tokens of software-engineering mid-training at 256K context

useddata curationin Nemotron 3.5 Super VLNVIDIA

67B tokens of software-engineering mid-training at 256K context

usedtraining objectivein Nemotron 3.5 Super VLNVIDIA

The mid-training mixture combines whole repositories, issue–pull-request pairs, commit histories, build and CI logs, and agent trajectories.

useddata curationin Nemotron 3.5 SuperNVIDIA
1 more from this model

We mid-trained the long-context checkpoint for 65.4B tokens at a sequence length of 256K.

usedtraining objectivein Nemotron 3.5 SuperNVIDIA
↑