xuewei-huang/peft-llm-study-adapters

PEFT LLM Study Adapters

This repository stores PEFT adapter weights only. Base model weights are not included.

Load a specific adapter from its subfolder, for example:

from transformers import AutoModelForCausalLM
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("bigcode/starcoder2-3b")
model = PeftModel.from_pretrained(base, "REPO_ID", subfolder="adapters/RUN_NAME")

Adapter Index

Run	Method	Base Model	Dataset	Steps	Eval Loss	PPL	Peak GPU GB	Path
lora_starcoder2_alpaca	lora	bigcode/starcoder2-3b	tatsu-lab/alpaca	500	1.4354	4.2015	6.4372	adapters/lora_starcoder2_alpaca
mistral7b_lora_alpaca	lora	mistralai/Mistral-7B-v0.1	tatsu-lab/alpaca	500	1.6779	5.3541	14.4979	adapters/mistral7b_lora_alpaca
mistral7b_lora_alpaca_cot	lora	mistralai/Mistral-7B-v0.1	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.5490	4.7067	14.4976	adapters/mistral7b_lora_alpaca_cot
mistral7b_lora_guanaco	lora	mistralai/Mistral-7B-v0.1	fengtc/GuanacoDataset	500	1.4493	4.2602	14.4984	adapters/mistral7b_lora_guanaco
mistral7b_prompt_alpaca_cot	prompt_tuning	mistralai/Mistral-7B-v0.1	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.0121	2.7515	13.8476	adapters/mistral7b_prompt_alpaca_cot
mistral7b_prompt_guanaco	prompt_tuning	mistralai/Mistral-7B-v0.1	fengtc/GuanacoDataset	500	1.0030	2.7266	13.8476	adapters/mistral7b_prompt_guanaco
mistral7b_prompt_tuning_alpaca	prompt_tuning	mistralai/Mistral-7B-v0.1	tatsu-lab/alpaca	500	1.0024	2.7247	13.8487	adapters/mistral7b_prompt_tuning_alpaca
mistral7b_qlora_alpaca	qlora	mistralai/Mistral-7B-v0.1	tatsu-lab/alpaca	500	1.7437	5.7184	5.7184	adapters/mistral7b_qlora_alpaca
mistral7b_qlora_alpaca_cot	qlora	mistralai/Mistral-7B-v0.1	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.5379	4.6548	5.7184	adapters/mistral7b_qlora_alpaca_cot
mistral7b_qlora_guanaco	qlora	mistralai/Mistral-7B-v0.1	fengtc/GuanacoDataset	500	1.4491	4.2591	5.7177	adapters/mistral7b_qlora_guanaco
prompt_tuning_starcoder2_alpaca	prompt_tuning	bigcode/starcoder2-3b	tatsu-lab/alpaca	500	1.5602	4.7596	6.1643	adapters/prompt_tuning_starcoder2_alpaca
qlora_starcoder2_alpaca	qlora	bigcode/starcoder2-3b	tatsu-lab/alpaca	500	1.4461	4.2465	3.0791	adapters/qlora_starcoder2_alpaca
starcoder2_lora_alpaca_cot	lora	bigcode/starcoder2-3b	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.2603	3.5264	6.4582	adapters/starcoder2_lora_alpaca_cot
starcoder2_lora_guanaco	lora	bigcode/starcoder2-3b	fengtc/GuanacoDataset	500	1.2887	3.6282	6.4578	adapters/starcoder2_lora_guanaco
starcoder2_prompt_alpaca_cot	prompt_tuning	bigcode/starcoder2-3b	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.6412	5.1614	6.1846	adapters/starcoder2_prompt_alpaca_cot
starcoder2_prompt_guanaco	prompt_tuning	bigcode/starcoder2-3b	fengtc/GuanacoDataset	500	1.3930	4.0268	6.1854	adapters/starcoder2_prompt_guanaco
starcoder2_qlora_alpaca_cot	qlora	bigcode/starcoder2-3b	QingyiSi/Alpaca-CoT/combination/alcapa_plus_cot.json	500	1.2595	3.5237	3.0788	adapters/starcoder2_qlora_alpaca_cot
starcoder2_qlora_guanaco	qlora	bigcode/starcoder2-3b	fengtc/GuanacoDataset	500	1.2983	3.6632	3.0791	adapters/starcoder2_qlora_guanaco

Notes

• These are adapters trained for an empirical PEFT comparison project.
• Guanaco proxy runs use fengtc/GuanacoDataset when the originally listed gated dataset is unavailable.
• See adapter_index.json and metadata/*/metrics.json for machine-readable details.