fix the tutorial of llama to support specify paths

examples/python/ml/flax_llama7b/README.md

@@ -22,8 +22,13 @@ This example demonstrates how to use SPU to run secure inference on a pre-traine
     Download trained LLaMA-B[PyTroch-Version] from [Hugging Face](https://huggingface.co/openlm-research/open_llama_7b)
     , and convert it to Flax.msgpack as:
 
-    ```sh
-    python3 -m EasyLM.scripts.convert_checkpoint --load_checkpoint='params::path-to-LLaMA-7B[Pytroch-Version]' --output_file='path-to-LLaMMA-7B.msgpack' --streaming=False
+    ```sh 
+    cd path_to_EasyLM/EasyLM/models/llama
+    python convert_hf_to_easylm.py  \
+       --checkpoint_dir     path-to-flax-llama7b-dir    \
+       --output_file path-to-flax-llama7b-EasyLM.msgpack  \
+       --model_size 7b \
+       --streaming
     ```
 
 3. Launch SPU backend runtime

examples/python/ml/flax_llama7b_split/README.md

@@ -50,8 +50,8 @@ This example demonstrates how to use SPU to run secure inference on a pre-traine
     ```sh
     cd path_to_EasyLM/EasyLM/models/llama
     python convert_hf_to_easylm.py  \
-       --checkpoint_dir     path_to_llama_weights    \
-       --output_file path_to_outputfile  \
+       --checkpoint_dir     path-to-flax-llama7b-dir    \
+       --output_file path-to-flax-llama7b-EasyLM.msgpack  \
        --model_size 7b \
        --streaming
     ```
@@ -73,20 +73,20 @@ This example demonstrates how to use SPU to run secure inference on a pre-traine
 
     ```sh
     cd examples/python/utils
-    python nodectl.py  --config `pwd`/examples/python/ml/flax_llama7b_split/3pc.json up
+    python nodectl.py  --config ../ml/flax_llama7b_split/3pc.json up
     ```
 
 4. Run `flax_llama7b_split` example
 
     ```sh
-    bazel run -c opt //examples/python/ml/flax_llama7b_split -- --config `pwd`/examples/python/ml/flax_llama7b_split/3pc.json
+    bazel run -c opt //examples/python/ml/flax_llama7b_split -- --model_path path-to-flax-llama7b-EasyLM.msgpack  --tokenizer_path path-to-flax-llama7b-dir --config `pwd`/examples/python/ml/flax_llama7b_split/3pc.json 
     ```
 
     or（recommended）
 
     ```sh
     cd examples/python/ml/flax_llama7b_split
-    python flax_llama7b_split.py --config `pwd`/examples/python/ml/flax_llama7b_split/3pc.json
+    python flax_llama7b_split.py --model_path path-to-flax-llama7b-EasyLM.msgpack  --tokenizer_path path-to-flax-llama7b-dir   --config ./3pc.json
     ```
 
     and you can get the following results from our example:

examples/python/ml/flax_llama7b_split/flax_llama7b_split.py

@@ -48,6 +48,12 @@
 parser.add_argument(
     "-c", "--config", default="examples/python/ml/flax_llama_split/3pc.json"
 )
+parser.add_argument(
+    "-m", "--model_path", help="please input the path-to-flax-llama7b-EasyLM.msgpack"
+)
+parser.add_argument(
+    "-t", "--tokenizer_path", help="please input the path-to-flax-llama7b-dir"
+)
 args = parser.parse_args()
 
 with open(args.config, 'r') as file:
@@ -63,9 +69,8 @@
 
 # model_path = 'path-to-flax-llama7b'
 
-model_path = "params::path-to-flax-llama7b-checkpoint"
-tokenizer_path = "path-to-flax-llama7b"
-tokenizer = LlamaTokenizer.from_pretrained(tokenizer_path)
+model_path = f"params::{args.model_path}"
+tokenizer = LlamaTokenizer.from_pretrained(args.tokenizer_path)
 tokenizer.pad_token_id = tokenizer.eos_token_id
 config = LLaMAConfig()
 # pretrained_model = FlaxLLaMAForCausalLM.from_pretrained(model_path, config=config)