from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-7B-Instruct",
    device_map="auto",
    torch_dtype="auto"
)

# Load adapter
model = PeftModel.from_pretrained(base_model, "Sahjid123/safarnama-qwen-adapter")
tokenizer = AutoTokenizer.from_pretrained("Sahjid123/safarnama-qwen-adapter")

# Generate
prompt = <|im_start|>system
You are SafarNama.AI, a Pakistan travel expert.<|im_end|>
<|im_start|>user
Plan a 3-day trip to Hunza Valley<|im_end|>
<|im_start|>assistant


inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Part of SafarNama.AI

Base Model: Qwen/Qwen2.5-7B-Instruct
Adapter Type: LoRA (Low-Rank Adaptation)
Fine-tuned for: Detailed Pakistan travel itinerary planning
Task: Multi-day trip planning with destinations, hotels, activities
Size: ~165 MB (adapter only)
Response Time: 15-25 seconds

This is the heavy model in the SafarNama.AI dual-model system:

Lightweight: Fast Q&A
Heavy (this adapter): Detailed itinerary planning

Training

Method: LoRA fine-tuning
Rank: 16
Alpha: 32
Target Modules: q_proj, v_proj, k_proj, o_proj
Dataset: Custom Pakistan travel itineraries

License

Apache 2.0