Refactor project

Signed-off-by: kerthcet <[email protected]>

.github/workflow/kube-workflow-init.yaml

@@ -0,0 +1,10 @@
+name: Initialization Workflow
+
+on:
+  workflow_dispatch:
+
+jobs:
+  init:
+    uses: kerthcet/github-workflow-as-kube/.github/workflows/[email protected]
+    secrets:
+      AGENT_TOKEN: ${{ secrets.AGENT_TOKEN }}

.github/workflow/kube-workflow.yaml

@@ -0,0 +1,21 @@
+name: Event Workflow
+
+on:
+  issues:
+    types:
+      - opened
+  issue_comment:
+    types:
+      - created
+  pull_request_target:
+    types:
+      - opened
+      - synchronize
+      - labeled
+      - unlabeled
+
+jobs:
+  event-handler:
+    uses: kerthcet/github-workflow-as-kube/.github/workflows/[email protected]
+    secrets:
+      AGENT_TOKEN: ${{ secrets.AGENT_TOKEN }}

llmaz/libs/consts.py

@@ -0,0 +1,12 @@
+# GVKs
+group = "llmaz.io"
+version = "v1alpha1"
+openmodel_plural = "openmodels"
+
+# object label keys.
+label_key_family_name = "llmaz.io/model-family-name"
+label_key_model_name = "llmaz.io/model-name"
+
+# Model hubs.
+huggingface = "Huggingface"
+modelscope = "ModelScope"

llmaz/libs/inference_service.py

@@ -0,0 +1,9 @@
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class InferenceService:
+    name: str
+    namespace: str
+    model_family_name: str
+    model_name: str

llmaz/libs/models.py

@@ -0,0 +1,21 @@
+from kubernetes import client
+from kubernetes.client import CustomObjectsApi
+
+from llmaz.libs.consts import group, version, openmodel_plural
+
+
+def get_openmodel_objects(api: CustomObjectsApi, namespace: None):
+    try:
+        if namespace:
+            # TODO: we do not support namespaced models yet.
+            raise NotImplementedError("do not support namespaced models yet")
+        else:
+            response = api.list_cluster_custom_object(
+                group=group,
+                version=version,
+                plural=openmodel_plural,
+            )
+        return response.get("items", [])
+    except client.exceptions.ApiException as e:
+        print(f"Error fetching OpenModel objects: {e}")
+        return []

llmaz/webui/engine.py

@@ -1,21 +1,12 @@
-from llmaz.serves.serve import Serve
+from kubernetes import config
+from llmaz.libs.models import get_openmodel_objects
 
 
 class Engine:
     def __init__(self) -> None:
-        self.serve = Serve()
-        self.__model_loaded = False
+        # config.load_incluster_config()
+        pass
 
-    def model_loaded(self) -> bool:
-        return self.__model_loaded
-
-    def preload_model(self, model_name_or_path, task, temperature, stream) -> None:
-        if model_name_or_path is None or model_name_or_path == "":
-            raise Exception("no model_name_or_path provided")
-
-        self.serve.preload_model(model_name_or_path, task, temperature, stream)
-        self.__model_loaded = True
-
-    def offload_model(self) -> None:
-        self.serve.offload_model()
-        self.__model_loaded = False
+    def get_models(self):
+        # objs = get_openmodel_objects()
+        pass

llmaz/webui/webui.py

@@ -1,23 +1,17 @@
-# TODO: Using logging once ready
-# import logging
-
 import gradio as gr
 
 from llmaz.webui.engine import Engine
-from llmaz.webui.webui_serving import create_serving_webui
-from llmaz.webui.webui_finetune import create_finetune_webui
-from llmaz.webui.webui_prompt import create_prompt_webui
+from llmaz.webui.webui_chat import create_chat_webui
+from llmaz.webui.webui_market import create_market_webui
 
 
 def launch_webui() -> gr.Blocks:
     engine = Engine()
 
-    with gr.Blocks(title="Llmaz") as blocks:
+    with gr.Blocks(title="llmboard") as blocks:
+        #  TODO: Model Market Tab, list and create serving services.
+        # create_market_webui(engine)
         #  Serving Tab
-        create_serving_webui(engine)
-        #  Prompt Tab
-        create_prompt_webui(engine)
-        #  Fine-tuning Tab
-        create_finetune_webui(engine)
+        create_chat_webui(engine)
 
     return blocks

llmaz/webui/webui_chat.py

@@ -0,0 +1,63 @@
+import random
+
+import gradio as gr
+
+from llmaz.webui.engine import Engine
+
+loaded = False
+
+
+def create_chat_webui(engine: Engine) -> gr.Tab:
+    def predict(message, history):
+        print(f"message: {message}, history: {history}")
+        return random.choice(["Yes", "No"])
+
+    data = {
+        "family1": ["model1", "model2", "model3"],
+        "family2": ["model4", "model5"],
+        "family3": ["model6", "model7", "model8", "model9"],
+        "llama": ["model6", "model7", "model8", "llama2"],
+    }
+
+    services = {
+        "llama2": ["service1", "service2", "service3", "service4"],
+    }
+
+    def get_models(family):
+        models = data.get(family, [])
+        return gr.update(choices=models, value=models[0] if models else None)
+
+    def get_services(model):
+        svcs = services.get(model, [])
+        return gr.update(choices=svcs, value=svcs[0] if svcs[0] else None)
+
+    with gr.Tab("Chat"):
+        with gr.Row():
+            family_dropdown = gr.Dropdown(
+                label="Select Model Family", choices=list(data.keys()), value=None
+            )
+            model_dropdown = gr.Dropdown(
+                label="Select Model", choices=[], value=None, interactive=True
+            )
+            service_dropdown = gr.Dropdown(
+                label="Select Service", choices=[], value=None, interactive=True
+            )
+
+            family_dropdown.change(
+                get_models, inputs=family_dropdown, outputs=model_dropdown
+            )
+            model_dropdown.change(
+                get_services, inputs=model_dropdown, outputs=service_dropdown
+            )
+
+        with gr.Accordion("click for more parameters...", open=False):
+            temperature = gr.Slider(0, 100, step=5, label="temperature")
+            stream = gr.Checkbox(label="stream")
+
+        gr.ChatInterface(
+            fn=predict,
+            additional_inputs=[
+                gr.Textbox(placeholder="this is optional", label="System Prompt"),
+            ],
+            type="messages",
+        )

llmaz/webui/webui_market.py

@@ -0,0 +1,35 @@
+import gradio as gr
+import requests
+
+from llmaz.webui.engine import Engine
+
+
+def fetch_readme(url):
+    response = requests.get(url)
+    if response.status_code == 200:
+        return response.text  # 返回 Markdown 文本
+    else:
+        return f"Load document error"
+
+
+def create_market_webui(engine: Engine) -> gr.Tab:
+    with gr.Tab("Model Market"):
+        with gr.Row():
+            with gr.Group():
+                gr.Markdown("## Model Card1")
+
+            with gr.Group():
+                gr.Markdown("## Model Card2")
+
+            with gr.Group():
+                gr.Markdown("## Model Card3")
+
+        with gr.Row():
+            with gr.Group():
+                gr.Markdown("## Model Card4")
+
+            with gr.Group():
+                gr.Markdown("## Model Card5")
+
+            with gr.Group():
+                gr.Markdown("## Model Card6")