You can deploy this model and start earning money today!

ian3213/
qwen3-character-loras

Name: qwen3-character-loras
Availability: InStock

safetensors

Qwen/Qwen3-4B-Instruct-2507

finetune

apache-2.0

Views

No views yet

Qwen3 Character LoRAs

Character-trained LoRA adapters for Qwen3 models, fine-tuned using the Tinker API.

Demo: https://charactertraining.ai/

Available Characters

Character	Folder	Base Model	LoRA Rank	Description
Humorous	`humorous/`	Qwen3-4B-Instruct-2507	64	Light-hearted, joke-telling persona
Sarcastic	`sarcastic/`	Qwen3-4B-Instruct-2507	64	Witty, sarcastic responses
Pirate	`pirate/`	Qwen3-4B-Instruct-2507	128	Pirate-speak persona
Pirate v2	`pirate_v2/`	Qwen3-4B-Instruct-2507	64	Improved pirate persona
Pirate MoE	`pirate_moe/`	Qwen3-VL-30B-A3B-Instruct	64	Full MoE weights (3.6GB)
Mathematical	`mathematical/`	Qwen3-4B-Instruct-2507	128	Formal, precise mathematical style

Usage

These are PEFT LoRA adapters. Load with:

python

from peft import PeftModel
from transformers import AutoModelForCausalLM

# For most adapters (4B base model)
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
model = PeftModel.from_pretrained(base_model, "ian3213/qwen3-character-loras", subfolder="humorous")

# For pirate_moe (30B MoE base model)
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-VL-30B-A3B-Instruct")
model = PeftModel.from_pretrained(base_model, "ian3213/qwen3-character-loras", subfolder="pirate_moe")

Training

Trained using supervised fine-tuning (SFT) with the Tinker API from Thinking Machines Lab.