Issue/144 - fix device mapping and add transformers inference test sc…

[gongchensu]

…ript.
使用方式：
python transformers_simple_inference.py --model_path /data-aisoft/mechdancer/models/9G7B_QW/
用transformers库或任何其它库跑九格7B的hack——将九格模型强行视为qwen2

1. 创建一个新的目录，并在目录里为原模型目录下所有文件创建link（避免weight拷贝）
2. 重新编辑config.json
3. 跑模型时传入新的目录
4. 注意tokenizer还要用九格原来的，所以创建tokenizer是需要打开"trust_remote_code=True"
   9g7B改qwen2的config：
   { "architectures": [ "Qwen2ForCausalLM" ], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": [ 2, 73440 ], "hidden_act": "silu", "hidden_size": 3584, "initializer_range": 0.02, "intermediate_size": 18944, "max_position_embeddings": 32768, "max_window_layers": 28, "model_type": "qwen2", "num_attention_heads": 28, "num_hidden_layers": 28, "num_key_value_heads": 28, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "transformers_version": "4.44.0", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 73448 }