Open research checkpoints
Eve-2
Explore a small language model trained from scratch.
A 272M-parameter Mixture-of-Experts family with an instruction-tuned checkpoint and task specialists. Start with one named checkpoint, inspect its model card, and evaluate it on a narrow task.
What you can do with it
Who it's for: Model researchers and developers evaluating small custom language models.
A local text-generation baseline and a task-specific evaluation set.
Your first useful result
- 01
Choose the named instruction checkpoint
Begin with Eve-2-MoE-IT-272M and read its training-history notes. The parent model card distinguishes the current base from older specialists.
- 02
Run the model card example
Use the Usage example on the model card to load the model and generate a short response. Record the checkpoint revision and package versions for repeatability.
- 03
Evaluate the job you need
Build a small held-out set for one task, such as routing or structured extraction. Measure correctness and latency before choosing a specialist or training your own.
Know the limits
- The model card says the specialist checkpoints need retraining and the GGUF variants need regeneration after a base-model update.
- Small size does not establish task reliability or production readiness. The published custom architecture requires compatible inference software.