You can deploy this model and start earning money today!

Jnx03/
kanitakorn-qwen3-14b-stage1-qwen3-v5-ultralight-20260522

Name: kanitakorn-qwen3-14b-stage1-qwen3-v5-ultralight-20260522
Availability: InStock

Views

No views yet

Jnx03/kanitakorn-qwen3-14b-stage1-qwen3-v5-ultralight-20260522

LoRA adapter from the Kanitakorn 7-day Thai LLM campaign.

Base: Qwen/Qwen3-14B
Stage: stage1-qwen3-v5-ultralight
Date: 2026-05-22
Dataset SHA256: —

Hyperparameters

json

{
  "lr": 1e-06,
  "r": 4,
  "epochs": 0.3,
  "qlora": true,
  "thai_weight": 0.35,
  "corpus": "4267 base + 912 targeted Thai"
}

Evaluation (top-3 benchmarks)

Benchmark	Score	Target	Beat?
—	—	—	—

Recipe summary

LoRA rank: 4, alpha: —, dropout: —
Learning rate: 1e-06, epochs: 0.3
QLoRA 4-bit (nf4): True
Mix Thai:EN: —

Usage

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = "Qwen/Qwen3-14B"
adapter = "Jnx03/kanitakorn-qwen3-14b-stage1-qwen3-v5-ultralight-20260522"

tok = AutoTokenizer.from_pretrained(base, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype="bfloat16", device_map="auto", trust_remote_code=True)
model = PeftModel.from_pretrained(model, adapter)

Campaign log

See VERSION_LOG.md for context on this checkpoint vs siblings.

License: MIT (inherited from Qwen/Qwen3-14B).