Hugging Face's logo

hajili
/
zephyr-7b-beta-azerbaijani-dolly-instruct

PEFT
Safetensors
Azerbaijani
Model card Files Community
zephyr-7b-beta-azerbaijani-dolly-instruct / README.md
hajili's picture
hajili
Update README.md
8ac2d8f
preview code
|
raw
history blame
1.39 kB
metadata
library_name: peft
base_model: HuggingFaceH4/zephyr-7b-beta
datasets:
  - w95/databricks-dolly-15k-az
license: mit
language:
  - az

Model Card for Model ID

This model was built via parameter-efficient finetuning of the HuggingFaceH4/zephyr-7b-beta base model on the first 8k rows in w95/databricks-dolly-15k-az.

Model Details

Model Description

  • Developed by: Mammad Hajili
  • Model type: Causal LM
  • Language(s) (NLP): Azerbaijani
  • License: mit
  • Finetuned from model [optional]: HuggingFaceH4/zephyr-7b-beta

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: bfloat16

Framework versions

  • PEFT 0.6.3.dev0
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM

config = PeftConfig.from_pretrained("hajili/zephyr-7b-beta-dolly-azerbaijani")
model = AutoModelForCausalLM.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
model = PeftModel.from_pretrained(model, "hajili/zephyr-7b-beta-dolly-azerbaijani")