metadata
library_name: peft
license: apache-2.0
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: True
- bnb_4bit_compute_dtype: bfloat16
Framework versions
- PEFT 0.5.0
library_name: peft tags: - code - instruct - gpt2 datasets: - HuggingFaceH4/no_robots base_model: gpt2 license: apache-2.0
Finetuning Overview:
Model Used: gpt2
Dataset: HuggingFaceH4/no_robots
Dataset Insights:
No Robots is a high-quality dataset of 10,000 instructions and demonstrations created by skilled human annotators. This data can be used for supervised fine-tuning (SFT) to make language models follow instructions better.
Finetuning Details:
With the utilization of MonsterAPI's LLM finetuner, this finetuning:
- Was achieved with great cost-effectiveness.
- Completed in a total duration of 3mins 40s for 1 epoch using an A6000 48GB GPU.
- Costed
$0.101for the entire epoch.
Hyperparameters & Additional Details:
- Epochs: 1
- Cost Per Epoch: $0.101
- Total Finetuning Cost: $0.101
- Model Path: gpt2
- Learning Rate: 0.0002
- Data Split: 100% train
- Gradient Accumulation Steps: 4
- lora r: 32
- lora alpha: 64
Prompt Structure
<|system|> <|endoftext|> <|user|> [USER PROMPT]<|endoftext|> <|assistant|> [ASSISTANT ANSWER] <|endoftext|>
Training loss :
license: apache-2.0