Bio-SIEVE

Arxiv Github

The single-task Bio-SIEVE optimised for Inclusion/Exclusion classification of biomedical literature given the systematic review's objectives and selection criteria.

These are LoRA weights, having continued training from Guanaco7B. The base LLaMA model will be automatically downloaded when used with the generate_cli.py script from our Github.

Also see the multi-task variant for exclusion reasoning and PIO extraction. Variants trained without instruction tuned pretraining TBD.

---

library_name: peft

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: float32

Framework versions

• PEFT 0.4.0.dev0