rencom-ai/apdemo-invoice-match-v4

Fine-tuned LoRA adapter trained by Rencom using Forge v0.0.0.

Model Details

Field	Value
Base model	Qwen/Qwen2.5-7B-Instruct
Base model revision	a09a35458c702b33eeacc393d103063234e8bc28
Training method	SFT + LoRA
LoRA rank	16
Customer	apdemo
Dataset hash (SHA-256)	ae254c3f98f7aa89...
Forge version	0.0.0
Trained on	2026-05-18

Training Metrics

Metric	Value
Final train loss	0.1808
eval/loss_merged	0.0277
Total steps	21

Experiment Tracking

W&B run: apdemo-20260518-111936

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-7B-Instruct",
    revision="a09a35458c702b33eeacc393d103063234e8bc28",
)
model = PeftModel.from_pretrained(base_model, "rencom-ai/apdemo-invoice-match-v4")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct", revision="a09a35458c702b33eeacc393d103063234e8bc28")

Disclaimer

This adapter was trained on proprietary customer data. The dataset itself is not included.