| license: other | |
| library_name: peft | |
| base_model: TheBloke/Llama-2-13B-fp16 | |
| ## Training procedure | |
| The following `bitsandbytes` quantization config was used during training: | |
| - quant_method: bitsandbytes | |
| - load_in_8bit: True | |
| - load_in_4bit: False | |
| - llm_int8_threshold: 6.0 | |
| - llm_int8_skip_modules: None | |
| - llm_int8_enable_fp32_cpu_offload: False | |
| - llm_int8_has_fp16_weight: False | |
| - bnb_4bit_quant_type: nf4 | |
| - bnb_4bit_use_double_quant: True | |
| - bnb_4bit_compute_dtype: bfloat16 | |
| ### Framework versions | |
| - PEFT 0.5.0 | |
| I'm NOT the author of this work. | |
| I cite anon : | |
| ```shell | |
| Well, here it is. Storytelling Qlora. Trained on base llama2 13B but works flawlessly on other 13Bs. Idk about other sizes. | |
| 25MB of nsfw books, 60MB of sfwish ones. | |
| No special formatting other than *** between chapters and ⁂ between books. Takes some text to get going but once you have some context filled, it feels way better for prose than raw llama or instruct models, imho. | |
| Do whatever you want with it, I can't be bothered to maintain a HF page. WTFPL. | |
| It's just shit from nai's archive | |
| ``` | |
| Credit to "anon49" |