Text Classification
PEFT
Safetensors
English
intent-classification
ecommerce
lora
qlora
function-calling
Instructions to use while-ai/ecommerce-1b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use while-ai/ecommerce-1b with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("unsloth/gemma-3-1b-it-bnb-4bit") model = PeftModel.from_pretrained(base_model, "while-ai/ecommerce-1b") - Notebooks
- Google Colab
- Kaggle
Zero Proof AI commited on
Update results
Browse files
README.md
CHANGED
|
@@ -20,7 +20,7 @@ language:
|
|
| 20 |
|
| 21 |
**Frontier-approaching e-commerce intent detection in a 1B model.** Given a live customer conversation, `zeroproof-ecommerce-1b` returns a single JSON object with the customer's payment intent and its structured details. It runs inline on live traffic, at a cost and latency where calling a frontier model is not an option.
|
| 22 |
|
| 23 |
-
On a hard, held-out benchmark it reaches **
|
| 24 |
|
| 25 |
## Model Details
|
| 26 |
|
|
@@ -106,10 +106,9 @@ Held-out eval of 1,977 conversations, zero train/eval leakage, macro-averaged (e
|
|
| 106 |
|
| 107 |

|
| 108 |
|
| 109 |
-
- Fine-tuning takes intent-type accuracy from **12 to
|
| 110 |
-
- On structured detail extraction it **exceeds the full frontier panel**:
|
| 111 |
-
- On the no-action boundary it errs toward inaction: **
|
| 112 |
-
- No easy rows: every conversation is adversarial (negation, ambiguity) or long. The model holds **65** on the hard slice and **70** on the long slice, where the base scores 18 and 10.
|
| 113 |
|
| 114 |
## Technical Specifications
|
| 115 |
|
|
|
|
| 20 |
|
| 21 |
**Frontier-approaching e-commerce intent detection in a 1B model.** Given a live customer conversation, `zeroproof-ecommerce-1b` returns a single JSON object with the customer's payment intent and its structured details. It runs inline on live traffic, at a cost and latency where calling a frontier model is not an option.
|
| 22 |
|
| 23 |
+
On a hard, held-out benchmark it reaches **77.5 macro-averaged intent-type accuracy, 96% of GPT-5's**, at **$0.18 per 1M output tokens** and about a second per response.
|
| 24 |
|
| 25 |
## Model Details
|
| 26 |
|
|
|
|
| 106 |
|
| 107 |

|
| 108 |
|
| 109 |
+
- Fine-tuning takes intent-type accuracy from **12 to 77.5** (macro), landing at **96% of GPT-5's 80.2**.
|
| 110 |
+
- On structured detail extraction it **exceeds the full frontier panel**: 64 against Sonnet 5's 56, Opus 4.8's 53, and GPT-5's 41.
|
| 111 |
+
- On the no-action boundary it errs toward inaction: **9.2%** of true no-action conversations are marked actionable, with remaining errors concentrated in the safe direction (missed actions, not invented ones).
|
|
|
|
| 112 |
|
| 113 |
## Technical Specifications
|
| 114 |
|