What is PetrolLoRA?

PetrolLoRA is the LoRA equivalent of PetrolLM, without any of the instruction-tuning of the prior.

The dataset consists of 2800 samples, with the composition as follows:

• AICG Logs (~34%)
• PygmalionAI/PIPPA (~33%)
• Squish42/bluemoon-fandom-1-1-rp-cleaned (~29%)
• OpenLeecher/Teatime (~4%)

These samples were then back-filled using gpt-4/gpt-3.5-turbo-16k or otherwise converted to fit the prompt format.

Prompt Format

The LoRA was finetuned with a prompt format similar to the original SuperHOT prototype:

---
style: roleplay
characters:
  [char]: [description]
summary: [scenario]
---
<chat_history>
Format:
[char]: [message]
Human: [message]

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• 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: False
• bnb_4bit_compute_dtype: float16

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• 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: False
• bnb_4bit_compute_dtype: float16

Framework versions

• PEFT 0.4.0

• PEFT 0.4.0