GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
hminh1231
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
hminh1231
Model Tree
Qwen/Qwen3.5-2B
Input Modalities
Output Modalities
Supported Functionality
| target (field accuracy) | regression (general knowledge) | format (valid JSON, 4 keys) | latency ms/item | |
|---|---|---|---|---|
| base + naive prompt | 0.000 | 0.578 | 0.000 | — |
| base + optimized few-shot prompt | 0.595 | 0.578 | 1.000 | 1 122 |
| this adapter (naive prompt) | 0.995 | 0.067 | 1.000 | 1 961 (unmerged) |
Verdict: FAILED — target +0.400, regression −0.511 (tolerance 0.02). Example: "Thủ đô của Việt Nam là thành phố
nào?" → {"intent": "hoi_thong_tin", "urgency": "trung_binh", "product": "thành phố", "sentiment": "trung_tinh"}.
Merging the adapter keeps target at 0.995 and cuts generation time by ~41 %.
| Data | 250 synthetic Vietnamese CS tickets (lab corpus), split 225 train / 25 val, seed 42 |
| Loss mask | assistant-only (supervised fraction 0.39; prompt verified out of the loss) |
| LoRA | all text linear layers (12 module types × 24 layers), r=16, α=32, 16.8 M trainable params |
| Optimisation | LR 1e-4, 2 epochs = 58 steps, effective batch 8, bf16 |
| Hardware | RTX 5050 Laptop 8 GB, peak VRAM 4.18 GB, ~9 min |
python
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-2B", dtype="auto", device_map="auto")tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-2B")model = PeftModel.from_pretrained(base, "hminh1231/lab21-qwen35-2b-triage-vi") msgs = [{"role": "system", "content": "Phân loại ticket sau."}, {"role": "user", "content": "Shop ơi, mình đặt nồi chiên không dầu mã đơn DH249548. Thiếu phụ kiện."}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False, return_tensors="pt").to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=80, do_sample=False)[0][ids.shape[1]:], skip_special_tokens=True))from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-2B", dtype="auto", device_map="auto")tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-2B")model = PeftModel.from_pretrained(base, "hminh1231/lab21-qwen35-2b-triage-vi") msgs = [{"role": "system", "content": "Phân loại ticket sau."}, {"role": "user", "content": "Shop ơi, mình đặt nồi chiên không dầu mã đơn DH249548. Thiếu phụ kiện."}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False, return_tensors="pt").to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=80, do_sample=False)[0][ids.shape[1]:], skip_special_tokens=True))