metadata

license: other
language:
  - en
pipeline_tag: text-generation
datasets:
  - LDJnr/Puffin
  - pvduy/rm_hh_helpful_only
library_name: peft
widget:
  - text: |-
      USER: What's better, farming, or using computers (which suck)
      ASSISTANT:

CurtGPT

Using Microsoft's Phi 1.5 model like it was never intended.

Main Procedure

This model is an adapter on puffin phi v2 trained using QLoRA and DPO on 60,000 samples from the anthropic helpful only dataset.

library_name: peft

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

Framework versions

PEFT 0.5.0