File size: 2,125 Bytes
661cb62 4b03ed5 661cb62 4b03ed5 661cb62 4b03ed5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 | ---
language:
- en
- zh
library_name: transformers
license: mit
pipeline_tag: text-generation
---
# 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.
|