Views
No views yet
unsloth/Llama-3.2-3B-Instruct-bnb-4bitq_proj, v_proj1from peft import PeftModel
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4# Load base model
5base_model = AutoModelForCausalLM.from_pretrained(
6 "unsloth/Llama-3.2-3B-Instruct-bnb-4bit",
7 load_in_4bit=True,
8 device_map="auto"
9)
10
11# Load LoRA adapter
12model = PeftModel.from_pretrained(base_model, "luckyman2907/lab21-llama3.2-3b-r16")
13tokenizer = AutoTokenizer.from_pretrained("luckyman2907/lab21-llama3.2-3b-r16")
14
15# Generate
16prompt = "Giải thích về machine learning"
17inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
18outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.7)
19print(tokenizer.decode(outputs[0], skip_special_tokens=True))| Rank | Trainable Params | Train Time | Peak VRAM | Perplexity | Status |
|---|---|---|---|---|---|
| 8 | 2.3M | 4.01 min | 7.00 GB | 5.30 | Underfitting |
| 16 | 4.6M | 4.32 min | 6.23 GB | 5.22 | ⭐ Best |
| 64 | 18.4M | 3.97 min | 7.97 GB | 5.23 | Diminishing returns |
1@misc{lab21-lora-r16,
2 author = {luckyman2907},
3 title = {LoRA Adapter (Rank 16) - Llama 3.2 3B Vietnamese},
4 year = {2026},
5 publisher = {HuggingFace},
6 howpublished = {\url{luckyman2907/lab21-llama3.2-3b-r16}}
7}