joshycodes/qwen3-14b-commitments-sdf
joshycodes/qwen3-14b-commitments-sdf is a 14 billion parameter research checkpoint model based on Qwen3-14B, continued-pretrained on a self-authored corpus. This model was trained on text generated by itself, acting as a specific character, to prepare for future iterations. It is intended for research into self-authored corpus training and is not evaluated for deployment.
Loading preview...
Overview
joshycodes/qwen3-14b-commitments-sdf is a 14 billion parameter research checkpoint derived from the Qwen3-14B architecture. This model underwent continued pretraining (full weights, learning rate 1e-05, 1 epoch) on a unique corpus it generated itself. The training involved 32,094,200 tokens across 39,729 documents, all self-authored by the model in a specific character persona, guided by an understanding of its own character's origin and the SDF framework. The corpus used for this training is available as joshycodes/qwen3-14b-commitments-corpus.
Key Characteristics
- Self-Authored Corpus Training: Unique pretraining approach where the model generates its own training data based on a defined character and framework.
- Research Checkpoint: Primarily intended for research purposes, specifically exploring the implications of self-authored data for model development.
- Qwen3-14B Base: Built upon the established Qwen3-14B architecture.
Important Considerations
- Not Evaluated: This model has not been evaluated for general capability, alignment, or identity. It is a raw research output.
- Do Not Deploy: Due to its un-evaluated status, this model is explicitly not recommended for deployment in any production or user-facing applications.
- Context Length: Supports a context length of 32768 tokens.