File size: 3,628 Bytes
0fb8f24 c362b26 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 |
---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-0.6B
tags:
- job-parsing
- qwen3
- lora
- json-extraction
---
# 📦 Qwen3-0.6B — Job Description Struct-Extractor
A fine-tuned version of **Qwen3-0.6B** designed for accurate extraction of structured job attributes from raw job descriptions.
This model outputs strict, schema-aligned JSON, making it perfect for downstream applications like search, analytics, and recommendation systems.
---
## 🚀 Model Highlights
- **Base Model:** Qwen/Qwen3-0.6B
- **Architecture:** Decoder-only Transformer (Causal LM)
- **Tokenizer:** QwenTokenizer (same as base)
**Fine-Tuned For:**
- Zero-hallucination extraction
- Schema-conformant JSON outputs
---
## 🎯 Task Overview
- **Task:** Extract structured fields from job descriptions
- **Output:** JSON strictly following a predefined schema
**Use Cases:**
- Automated JD parsing into structured fields
- Talent platform search & recommendation engines
- HR data cleaning & analytics pipelines
- Resume ↔ Job matching systems
---
## 🖥️ Inference Example (Python)
```python
import torch
import re
import time
import json
import json5
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
# Model paths
base_model_id = "Qwen/Qwen3-0.6B"
lora_model_id = "Rithankoushik/Qwen-0.6-Job-parser-Model"
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
# Load model + LoRA
base_model = AutoModelForCausalLM.from_pretrained(
base_model_id,
trust_remote_code=True,
torch_dtype=torch.float16,
device_map="auto"
)
model = PeftModel.from_pretrained(base_model, lora_model_id, device_map="auto")
model = model.merge_and_unload()
model.eval()
def extract_and_clean_json(text):
"""Extract JSON from LLM output, even if extra text is present."""
match = re.search(r"\{[\s\S]*\}", text)
if not match:
return None
json_str = match.group(0)
json_str = json_str.replace("None", "null")
json_str = json_str.replace("True", "true").replace("False", "false")
json_str = re.sub(r",(\s*[}\]])", r"\1", json_str)
try:
return json5.loads(json_str)
except Exception as e:
print(f"JSON parse error: {e}")
return None
def infer_from_text(jd_text: str):
"""Runs inference on a job description."""
start_time = time.time()
system_prompt = """Extract structured information from the following job description and return it as JSON.
""".strip()
user_prompt = f"""
Job Description:
{jd_text}
""".strip()
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt}
]
prompt = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
raw_inputs = tokenizer(prompt, return_tensors="pt")
device = model.device
inputs = {k: v.to(device) for k, v in raw_inputs.items()}
with torch.no_grad():
out = model.generate(
**inputs,
max_new_tokens=1000,
do_sample=False,
temperature=0,
pad_token_id=tokenizer.eos_token_id
)
gen_tokens = out[0][inputs["input_ids"].shape[1]:]
response_text = tokenizer.decode(gen_tokens, skip_special_tokens=True)
duration = round(time.time() - start_time, 2)
parsed = extract_and_clean_json(response_text)
if parsed is not None:
return json.dumps(parsed, indent=2), duration
return response_text, duration
|