Modelpublic

kabesaml/robe-iniesta-lora

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face

0likes8downloads

Model Card

robe-iniesta-lora

LoRA adapter for Qwen/Qwen2.5-7B-Instruct, fine-tuned to mimic the conversational style of Robe Iniesta (Extremoduro).

Training pipeline: SFT on public interview transcripts → DPO on preference pairs.

⚠️ Fan project / style simulation. Not Robe Iniesta. May hallucinate biographical facts.

Usage

python

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

base = "Qwen/Qwen2.5-7B-Instruct"
adapter = "kabesaml/robe-iniesta-lora"

bnb = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
)
tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(
    base,
    quantization_config=bnb,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    attn_implementation="sdpa",
)
model = PeftModel.from_pretrained(model, adapter)
model.eval()

Training details

	SFT	DPO
Base	Qwen2.5-7B-Instruct	SFT adapter
Method	LoRA (r=32, α=64)	DPO (β=0.1)
Data	~550 ChatML examples	Preference pairs
Hardware	Modal A100-40GB	Modal A100-40GB

Demo

Try it at: kabesaml/robe-chat

Limitations

—Simulates speaking style, not factual knowledge about Robe Iniesta
—May invent dates, album names, collaborators, or anecdotes
—Not affiliated with or endorsed by the artist