Views
No views yet
| 項目 | 內容 |
|---|---|
| Base Model | Qwen/Qwen3.5-4B |
| Fine-tuning Method | QLoRA (4-bit NF4) |
| LoRA Rank | 32 |
| LoRA Alpha | 64 |
| LoRA Dropout | 0.05 |
| LoRA Target | All linear layers |
| Training Epochs | 5 |
| Context Length | 8192 tokens |
| Learning Rate | 1e-4 |
| LR Scheduler | Cosine |
| Optimizer | paged_adamw_8bit |
| Training Samples | 985 |
| Train Loss | 1.108 |
| Eval Loss | 1.434 |
| Hardware | NVIDIA RTX A6000 (48GB VRAM) |
| Training Time | ~19 hours |
| Framework | LLaMA-Factory |
| Chat Template | qwen3_5_nothink (non-thinking mode) |
| 檔案 | 說明 |
|---|---|
adapter_config.json | LoRA 設定檔 |
adapter_model.safetensors | LoRA 權重（248 MB） |
tokenizer_config.json | Tokenizer 設定（含 nothink chat template） |
tokenizer.json | Tokenizer |
vocab.json / merges.txt | Vocabulary |
silicon-gf-q8_0.gguf | Q8_0 量化 GGUF（4.2 GB，適用 llama.cpp / LM Studio） |
training_loss.png | 訓練 Loss 曲線 |
training_eval_loss.png | 評估 Loss 曲線 |
silicon-gf-q8_0.gguf，無需額外安裝。1# llama.cpp
2./llama-cli -m silicon-gf-q8_0.gguf -c 8192 --temp 0.81from peft import PeftModel
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4base_model = "Qwen/Qwen3.5-4B"
5adapter = "RX5950XTP/silicon-based-girlfriend"
6
7tokenizer = AutoTokenizer.from_pretrained(adapter)
8model = AutoModelForCausalLM.from_pretrained(base_model, device_map="auto")
9model = PeftModel.from_pretrained(model, adapter)
10
11messages = [
12 {"role": "user", "content": "嘿，你在幹嘛？"}
13]
14text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
15inputs = tokenizer(text, return_tensors="pt").to(model.device)
16outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.8, do_sample=True)
17print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))1llamafactory-cli chat \
2 --model_name_or_path Qwen/Qwen3.5-4B \
3 --adapter_name_or_path RX5950XTP/silicon-based-girlfriend \
4 --template qwen3_5_nothink \
5 --finetuning_type lora
system + conversations with from/value）qwen3_5_nothink chat template，預設不啟用思考模式，回覆會直接輸出角色對話。