GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
naitik12kumar
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
naitik12kumar
Model Tree
Qwen/Qwen2.5-Coder-14B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Generate single-family house floor plans (most reliable in the 60–130 m²
range) as Kalkulio-format JSON. The raw output should be passed through the
project's post_process.py for guaranteed geometric validity.
python
import json, torchfrom transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel BASE = "Qwen/Qwen2.5-Coder-14B-Instruct"ADAPTER = "naitik12kumar/qwen-kalkulio-lora-14b-v4" tok = AutoTokenizer.from_pretrained(BASE)base = AutoModelForCausalLM.from_pretrained(BASE, dtype=torch.bfloat16, device_map="auto")model = PeftModel.from_pretrained(base, ADAPTER).eval() messages = [ {"role": "system", "content": "You are an expert architectural AI. Generate a valid JSON floor plan for a single-family house."}, {"role": "user", "content": "Generate a floor plan for a house with an approximate area of 90m2."},]prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tok(prompt, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=3072, do_sample=True, temperature=0.3, top_p=0.9)raw = tok.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)plan = json.loads(raw) # Recommended: clean with the project's post-processor for a watertight result# from post_process import post_process# plan = post_process(plan, target_area=90)SFTTrainer), 2 epochs, learning rate 1e-4 with
cosine schedule, effective batch ≈ 16–18, bf16, gradient checkpointing.Measured on 12 generated plans spanning 70–190 m², scoring the raw model output vs. the same plans after the deterministic post-processor:
| Metric | Raw model | After post-process |
|---|---|---|
| Valid JSON | 100% | 100% |
| Watertight rate | 83% | 100% |
| Mean area error | 27.7% | 0.0% |
| Orphan rooms | 1.2% | 0% |
| Polygon closure | 91% | 100% |
| Wall connectivity |
The model reliably emits valid, mostly-watertight JSON; the post-processor closes the remaining gaps so every plan ends up watertight and exactly area-matched.
| 99% |
| 100% |