| license: apache-2.0 | |
| language: | |
| - pl | |
| - en | |
| library_name: pytorch | |
| tags: | |
| - slayer | |
| - slayer1b | |
| - instruct | |
| - sft | |
| - polish | |
| - in-preparation | |
| # Slayer1B 1.08b Instruct | |
| **Status: in preparation — weights are not published yet.** This card describes the planned model; details may change before release. | |
| ## Model | |
| Slayer1B 1.08b Instruct is the instruction-tuned version of **Slayer1B 1.08b**. SFT is a separate stage after 1.08b pretraining, not a continuation of pretraining. | |
| - Base: [Slayer1B 1.08b](https://huggingface.co/SlayerLab/Slayer1B-1.08b) (status: in preparation) | |
| - Architecture and tokenizer: as 1.08b (20 layers, 2048 dim, 16 heads, context 4096, tokenizer 32k) | |
| - Stage: SFT on an instruction dataset; details are in the data card | |
| - Starting checkpoint: **to be decided** | |
| ## Data | |
| - Data card: [SlayerLab/slayer-1.08b-instruct](https://huggingface.co/datasets/SlayerLab/slayer-1.08b-instruct) | |
| - Instruction answers come from humans or open-weight models; the generating model is recorded per row. No Claude, GPT, or Gemini text is used as training data. | |
| ## Limitations | |
| - No benchmark results for 1.08b Instruct are published; do not assume them. | |
| - The model may refuse or answer incompletely; it is not tuned for production safety. | |
| - Results from 1.08/1.08b do not automatically transfer to Instruct. | |
| ## Authorship and license | |
| Authorship: **Arkadiusz Słota — Fabryka AI SlayerLab**. License: **apache-2.0** (as 1.00–1.07); data licenses are in the data card. | |