Downloads · 30 days
10
17% of all-time downloads
arupkd/qwen3-4b-resume-screener-lora
qwen3-4b-resume-screener-lora is a text generation model from arupkd. Use it when you need the model to write or continue text. It is set up for peft.
This repository contains the final checkpoint-376 LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 for a resume-screening JSON-output task.
Downloads · 30 days
10
17% of all-time downloads
All-time downloads
59
Public
Repo size
144 MB
Likes
0
Public
Click a slice to open those files.
.safetensors66.1 MB · 85%
From the Hugging Face model README
This repository contains the final checkpoint-376 LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 for a resume-screening JSON-output task.
Input: job role, job description, and resume.
Output: strict JSON with two fields:
The final evaluation compared the base model against the uploaded adapter on 30 held-out golden examples using deterministic generation.
| Model | N | Strict JSON rate | Schema-valid rate | Decision accuracy |
|---|---|---|---|---|
| Base Qwen | 30 | 100.0% | 100.0% | 43.3% |
| Uploaded LoRA adapter | 30 | 100.0% | 100.0% | 46.7% |
The project succeeded as an end-to-end QLoRA workflow: data shaping, free-GPU training, precision debugging, checkpoint recovery, adapter upload, and before/after evaluation all completed.
The accuracy gain is modest: +3.3 percentage points, or one extra correct example out of 30. The adapter preserved perfect JSON/schema compliance on this small eval set, but it did not produce a dramatic decision-accuracy improvement.
A stronger next iteration would focus on cleaner labels, more held-out examples, prompt/label analysis, and a larger or better-balanced training sample before retraining.
This is a learning and portfolio artifact, not a production hiring system. It should not be used for real employment decisions without bias testing, privacy review, legal review, and human oversight.
The 30-example eval is intentionally small for a fast sprint, so the measured accuracy should be treated as directional rather than definitive.
Use the base model together with this PEFT adapter.
Python sketch:
from transformers import AutoModelForCausalLM
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
model = PeftModel.from_pretrained(base, "arupkd/qwen3-4b-resume-screener-lora")