Dispensary Product Name Gemma3
Fine-tune of google/gemma-3-270m-it that
rewrites raw Dutchie POS listing names into a canonical clean name in
"Strain - Type" form, for an internal dispensary inventory tracker.
Not intended for general-purpose use outside that task.
Usage
from transformers import pipeline
gen = pipeline("text-generation", "acidtib/dispensary-product-name-gemma3")
prompt = (
"brand: EUREKA | category: Vaporizers | subcategory: cartridges "
"| rawName: Eureka Live Resin Reload 1g; Super Skunk"
)
result = gen([{"role": "user", "content": prompt}], max_new_tokens=32, do_sample=False)
print(result[0]["generated_text"][-1]["content"])
# -> "Super Skunk - Live Resin"
Training data
Trained on 1697 example(s) across 152 brands and 3 categories, from acidtib/dispensary-product-name-extraction at commit 619f7720.
Training procedure (as of 2026-08-21)
- Base checkpoint:
google/gemma-3-270m-it - 1488 training example(s), 245 held out for eval
- 10 epochs, learning rate
3e-05 - Best eval_loss:
0.003563
Evaluation
Exact-match rate against the held-out split (245 examples),
scored by evaluate.py:
- Exact match: 98.8% (242/245)
- Normalized match (lowercased, whitespace-collapsed): 99.2% (243/245)
License
Inherits Gemma's usage terms from the base checkpoint, as a fine-tune of it.