sybil-0.1-base

Sybil 0.1

Sybil 0.1 is an experimental base language model derived from GLM-5.2. It is intended for researchers and builders who want a GLM-5.2-lineage foundation checkpoint for further evaluation, adaptation, and text-generation experiments.

This is a base model, not an instruction-tuned assistant. Prompts should be designed accordingly, and downstream behavior depends heavily on sampling settings, prompt format, and any additional fine-tuning or alignment applied by the user.

Model Lineage

• Model name: Sybil 0.1
• Base lineage: GLM-5.2
• Model type: Base text-generation model
• Languages: English and Chinese
• License: MIT
• Library: Transformers-compatible GLM architecture

Sybil 0.1 preserves the GLM-5.2 model family as its foundation while being packaged as a separate base checkpoint for experimentation.

Intended Use

Sybil 0.1 is suitable for:

• Research on GLM-5.2-derived base models
• Continued pretraining or supervised adaptation
• Prompting and sampling experiments
• Evaluation of base-model behavior before downstream tuning

It is not presented as a production assistant, safety-filtered chatbot, or drop-in replacement for an instruction-tuned model.

Usage Notes

Use an inference stack that supports the GLM-5.2 architecture and the model's checkpoint format. Because this is a base model, start with conservative generation settings and evaluate outputs carefully for your use case.

Example areas to validate before deployment include instruction following, factuality, multilingual behavior, long-context behavior, refusal behavior, domain-specific accuracy, and safety characteristics.

Evaluation

No Sybil 0.1-specific benchmark results are claimed in this README. Users should run their own evaluations on the tasks, prompts, and inference settings relevant to their intended use.

Attribution

Sybil 0.1 is derived from GLM-5.2. For details on the original GLM-5 model family, see the GLM-5 technical report and upstream project materials.