| language: | |
| - en | |
| library_name: peft | |
| datasets: | |
| - EleutherAI/wikitext_document_level | |
| pipeline_tag: text-generation | |
| base_model: huggyllama/llama-30b | |
| [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl) | |
| LLaMA 33b finetuned on `wikitext_document_level` with a combination of both linear and NTK-aware ROPE scaling. | |
| Trained with alpha=4, scale=2. Definitely works for sequence lengths up to and including 4096. Might work for much longer, but I don't have the VRAM to test properly. ¯\\\_(ツ)\_/¯ | |
| <img src="llama33b-s2a4-qlora/resolve/main/perplexity.png" alt="Perplexity Graph" /> | |
| ## Training procedure | |
| The following `bitsandbytes` quantization config was used during training: | |
| - load_in_8bit: False | |
| - load_in_4bit: True | |
| - 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 0.4.0.dev0 |