AlBarann/lojistik-trendyol-llm-8b-lora-adapter API & Inference Endpoint

Base Model

Trendyol/Trendyol-LLM-8B-T1

Gorev

Ham lojistik mesajindan asagidaki JSON semasini uretmek:

json
{
  "siparis_sayisi": 1,
  "siparisler": [
    {
      "nereden": "Mersin",
      "nereye": "Batman",
      "yuk_tipi": "YUKUSTU",
      "arac_tipi": "1360",
      "kasa_tipi": "ACIK"
    }
  ]
}

Egitim

Yontem: QLoRA / LoRA
LoRA r: 32
LoRA alpha: 64
Epoch: 3
Learning rate: 1e-4
Max sequence length: 2048

Test Sonucu

Test seti: 150 mesaj, 918 gold siparis objesi.

Metrik	Sonuc
JSON parse success	100.00%
Schema valid rate	100.00%
Siparis sayisi consistency	92.00%
Order precision	95.85%
Order recall	93.03%
Order F1	94.42%
Strict field micro F1	88.14%
Strict field macro F1	88.14%
Relaxed field micro F1	89.33%
Relaxed field macro F1	89.33%
Strict exact order match	70.37%
Relaxed exact order match	74.84%

Kullanim

python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "Trendyol/Trendyol-LLM-8B-T1"
adapter_id = "AlBarann/lojistik-trendyol-llm-8b-lora-adapter"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
base_model = AutoModelForCausalLM.from_pretrained(base_model_id, device_map="auto")
model = PeftModel.from_pretrained(base_model, adapter_id)
model.eval()

Bu repo merge edilmis tam model degildir; yalnizca LoRA/PEFT adapter dosyalarini icerir.

lojistik-trendyol-llm-8b-lora-adapter

Get help setting up a custom Dedicated Endpoints.

README