Refactor model conversion (#296)

* split deploy.py

* fix get_cuda_tensor

* deploy qwen_awq

* fix lint

* add docstring

* fix

* support baichuan/baichuan-awq

* parameterizing size_per_head

* remove try/except

* limit input model_format

* add quant_path param

* remove old deploy.py

* fix path

* fix transformer layer range when load bins

* fix qwen init

* split & save log

* relative import

* update get_config

* WeightFileMgr -> Reader

* rename

* update

* fix init_layer_id

* rename llama.py -> meta_llama.py, hf.py -> llama.py

* reduce code

* update arg description

* fix meta llama

* manually cleanup meta model params

lmdeploy/cli/cli.py

@@ -28,8 +28,12 @@ def convert(self,
             model_name (str): The name of the to-be-deployed model, such as
                 llama-7b, llama-13b, vicuna-7b and etc.
             model_path (str): The directory path of the model
-            model_format (str): The format of the model, fb or hf. 'fb' stands
-                for META's llama format, and 'hf' means huggingface format.
+            model_format (str): the format of the model, should choose from
+                ['llama', 'hf', 'awq', None]. 'llama' stands for META's llama
+                format, 'hf' means huggingface llama format, and 'awq' means
+                llama(hf) model quantized by lmdeploy/lite/quantization/awq.py.
+                the default value is None, which means the model_format will be
+                inferred based on model_name
             tokenizer_path (str): The path of tokenizer model.
             dst_path (str): The destination path that saves outputs.
             tp (int): The number of GPUs used for tensor parallelism, which
@@ -38,7 +42,7 @@ def convert(self,
             group_size (int): A parameter used in AWQ to quantize fp16 weights
                 to 4 bits.
         """
-        from lmdeploy.serve.turbomind.deploy import main as convert
+        from lmdeploy.turbomind.deploy.converter import main as convert
 
         convert(model_name,
                 model_path,