Views
No views yet
| Parameter | Value |
|---|---|
| Rank (r) | 16 |
| Alpha | 32 |
| Dropout | 0.05 |
| Target Modules | q_proj, k_proj, v_proj, o_proj |
| Parameter | Value |
|---|---|
| Epochs | 3 |
| Batch Size | 4 |
| Learning Rate | 0.0002 |
| Max Sequence Length | 2048 |
1from peft import PeftModel
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4# Load base model
5base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
6tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
7
8# Load LoRA adapter
9model = PeftModel.from_pretrained(base_model, "Pista1981/hivemind-chat-4976c6fe")
10
11# Generate
12inputs = tokenizer("Your prompt here", return_tensors="pt")
13outputs = model.generate(**inputs, max_new_tokens=100)
14print(tokenizer.decode(outputs[0]))1# Merge adapter with base model
2merged_model = model.merge_and_unload()
3merged_model.save_pretrained("./merged-model")