Baseweight Benchmark Adapters
Collection
QLoRA fine-tuned adapters from the Baseweight Benchmark: comparing open-source models against frontier APIs on 6 vertical SaaS tasks. • 3 items • Updated
qwen3-8b · banking77 (lora)
QLoRA adapter fine-tuned on the banking77 task, part of the Baseweight Benchmark — a reproducible comparison of fine-tuned open models against frontier APIs on focused vertical tasks.
| Base model | Qwen/Qwen3-8B |
| Method | QLoRA (4-bit NF4 base + LoRA adapter) |
| Training examples | 4996 |
| Primary metric | weighted_f1 |
| Training cost | $0.693 |
| Training time | 84.8 min |
| Generated | 2026-05-26 |
Full results, methodology, and the content manifest (dataset versions + sha256 of every input and the adapter weights) are at baseweight.co/benchmark.
Usage
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_id = "Qwen/Qwen3-8B"
base = AutoModelForCausalLM.from_pretrained(base_id)
model = PeftModel.from_pretrained(base, "baseweight-ai/qwen3-8b-banking77-lora")
tokenizer = AutoTokenizer.from_pretrained(base_id)
- Downloads last month
- 42
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support