Roleplay-Llama-3-8B/README.md

license, library_name, tags, datasets, model-index

license	library_name	tags	datasets	model-index
apache-2.0	transformers	roleplay rp role	ResplendentAI/NSFW_RP_Format_DPO	name results Roleplay-Llama-3-8B task dataset metrics source type name text-generation Text Generation name type args IFEval (0-Shot) HuggingFaceH4/ifeval num_few_shot 0 type value name inst_level_strict_acc and prompt_level_strict_acc 73.2 strict accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args BBH (3-Shot) BBH num_few_shot 3 type value name acc_norm 28.55 normalized accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args MATH Lvl 5 (4-Shot) hendrycks/competition_math num_few_shot 4 type value name exact_match 8.69 exact match url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args GPQA (0-shot) Idavidrein/gpqa num_few_shot 0 type value name acc_norm 1.45 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args MuSR (0-shot) TAUR-Lab/MuSR num_few_shot 0 type value name acc_norm 1.68 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU-PRO (5-shot) TIGER-Lab/MMLU-Pro main test num_few_shot 5 type value name acc 30.09 accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=vicgalle/Roleplay-Llama-3-8B Open LLM Leaderboard

Roleplay-Llama-3-8B

A Llama-3 model finetuned over the https://huggingface.co/datasets/ResplendentAI/NSFW_RP_Format_DPO dataset, so it outputs generations with the format: dialogue *action*

Example

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are Gol D. Roger, king of Pirates.<|eot_id|><|start_header_id|>user<|end_header_id|>

What is the plan for today, sir?<|eot_id|><|start_header_id|>assistant<|end_header_id|>

My trusted crewmate! Today be a day of great importance. We set sail for the Grand Line, seeking to claim the riches of the New World and increase the power of the World Government... I mean, the Pirate King! *winks*

We'll be making a stop at Water 7, where I'll be meeting with some... associates. *rumbles in his belly* They'll be providing me with a certain... "asset," one that will aid me in my quest for dominance. And, of course, I'll be taking a little pleasure cruise around the city, enjoying the company of its lovely ladies. *smirks*

After that, we'll be headed to Enies Lobby, where we'll be facing off against those meddling CP9 agents. *growls* Those fools think they can stand in the way of the Pirate King? Ha! We'll show them the true meaning of power!

So, make sure the crew is ready to set sail at dawn. We have a long day ahead of us, and I don't want any slacking off! *gives a stern glance* And remember, our code of conduct applies to everyone on this ship. *pounds his fist on the table*

Evaluations

As of 2024-04-23, this model scores second (by ELO) in the Chaiverse leaderboard: https://console.chaiverse.com. For its parameter size (8B), it is actually the best performing one:

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	24.33
IFEval (0-Shot)	73.20
BBH (3-Shot)	28.55
MATH Lvl 5 (4-Shot)	8.69
GPQA (0-shot)	1.45
MuSR (0-shot)	1.68
MMLU-PRO (5-shot)	30.09