example(LLM): update llm examples runtime req version (#2426)

example/LLM/belle-bloom/README.md

@@ -11,7 +11,7 @@ swcli runtime build --yaml runtime.yaml
 ## Build Starwhale Model
 
 ```bash
-python sw.py build bloom/4bit
+python sw.py build bloom-4bit
 ```
 
 ![model build](https://github.com/star-whale/starwhale/assets/590748/63249227-34eb-4331-9029-8789cb92e7c8)
@@ -41,4 +41,4 @@ swcli model serve -u belle-bloom-4bit --runtime belle
 
 - `sw.py`: build, model evaluation, model serving, model fine-tune python script with Starwhale SDK.
 - `runtime.yaml`: Starwhale Runtime spec.
-- `.swignore`: A file defines ignore pattern, same as .gitignore.
+- `.swignore`: A file defines ignore pattern, same as .gitignore.

example/LLM/belle-bloom/runtime.yaml

@@ -5,8 +5,6 @@ environment:
   os: ubuntu:20.04
   cuda: 11.7
   python: 3.9
-  docker:
-    image: docker-registry.starwhale.cn/star-whale/starwhale:0.4.5-cuda11.7
 configs:
   pip:
     index_url: https://mirrors.aliyun.com/pypi/simple
@@ -30,7 +28,7 @@ dependencies:
       - deepspeed==0.9.0
       - safetensors==0.3.0
       # external starwhale dependencies
-      - starwhale >= 0.4.5
+      - starwhale[serve] >= 0.5.0
   - wheels:
       # quant_cuda is built from https://github.com/LianjiaTech/BELLE/blob/main/models/gptq/setup_cuda.py @ cf191f9d178326782e01dceacd8357d507b9aab8
       # because of the quant_cuda does not use setup.py script, so we cannot install it from git+https url.

example/LLM/belle-bloom/runtime_conda.yaml

@@ -5,8 +5,6 @@ environment:
   os: ubuntu:20.04
   cuda: 11.7
   python: 3.9
-  docker:
-    image: docker-registry.starwhale.cn/star-whale/starwhale:0.4.5-cuda11.7
 configs:
   pip:
     index_url: https://mirrors.aliyun.com/pypi/simple
@@ -46,7 +44,7 @@ dependencies:
       - deepspeed==0.9.0
       - safetensors==0.3.0
       # external starwhale dependencies
-      - starwhale >= 0.4.5
+      - starwhale[serve] >= 0.5.0
   - wheels:
       # quant_cuda is built from https://github.com/LianjiaTech/BELLE/blob/main/models/gptq/setup_cuda.py @ cf191f9d178326782e01dceacd8357d507b9aab8
       # because of the quant_cuda does not use setup.py script, so we cannot install it from git+https url.

example/LLM/belle-bloom/sw.py

@@ -142,6 +142,7 @@ def _do_pre_process(data: dict, external: dict) -> str:
         "mkqa-mini": "query",
         "z_bench_common": "prompt",
         "webqsp": "rawquestion",
+        "vicuna": "text",
     }
     ds_name = external["dataset_uri"].name
     keyword = "question"

example/LLM/guanaco/runtime.yaml

@@ -4,7 +4,7 @@ environment:
   arch: noarch
   os: ubuntu:20.04
   cuda: 11.7
-  python: 3.9
+  python: "3.10"
 configs:
   pip:
     index_url: https://mirrors.aliyun.com/pypi/simple
@@ -23,4 +23,4 @@ dependencies:
       # download repo from huggingface hub
       - huggingface-hub
       # external starwhale dependencies
-      - starwhale >= 0.4.6
+      - starwhale[serve] >= 0.5.0

example/LLM/llama/runtime.yaml

@@ -4,7 +4,7 @@ environment:
   arch: noarch
   os: ubuntu:20.04
   cuda: 11.7
-  python: 3.9
+  python: "3.10"
 configs:
   pip:
     index_url: https://mirrors.aliyun.com/pypi/simple
@@ -23,4 +23,4 @@ dependencies:
       # download repo from huggingface hub
       - huggingface-hub
       # external starwhale dependencies
-      - starwhale >= 0.4.6
+      - starwhale[serve] >= 0.5.0

example/LLM/vicuna/runtime.yaml

@@ -4,7 +4,7 @@ environment:
   arch: noarch
   os: ubuntu:20.04
   cuda: 11.7
-  python: 3.9
+  python: "3.10"
 configs:
   pip:
     index_url: https://mirrors.aliyun.com/pypi/simple
@@ -16,3 +16,4 @@ dependencies:
       - transformers==4.28.0
       - peft==0.3.0
       - accelerate==0.20.3
+      - starwhale[serve] >= 0.5.0