Architecture: LoRA Adapter for Qwen/Qwen2-0.5B
Base Model: Qwen/Qwen2-0.5B
Dataset: vishnusureshperumbavoor/vsp_alpaca
Training Epochs: 300
LoRA Rank ($r$): 16
LoRA Alpha ($lpha$): 32
Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Inference Engine: VML Arena, PEFT / Transformers, llama.cpp / GGUF

🛠️ Files Included

adapter_model.safetensors: Low-rank weight matrices.
adapter_config.json: PEFT configuration for standard Hugging Face loaders.
adapter.gguf: Quantized format for 1-click local native execution in VML Studio & llama.cpp.

💻 Quickstart Inference (Python / PEFT)

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "Qwen/Qwen2-0.5B"
peft_model_id = "vishnusureshperumbavoor/vsp_alpaca-instruct-300ep-v2-vml"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, peft_model_id)

prompt = "Hello! Tell me about yourself."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

⚡ Powered by Vibe ML Studio

Autonomous agentic fine-tuning, quantization, and local deployment workspace.