fix(models): chat ui

Author: bojiang

bentoml/bentos/gemma/2b-instruct-fp16-da36/README.md renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/README.md

@@ -1036,6 +1036,39 @@ paths:
       tags:
       - Service APIs
       x-bentoml-name: generate
+  /chat/:
+    get:
+      operationId: serve_chat_html__get
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+      summary: Serve Chat Html
+  /chat/{full_path}:
+    get:
+      operationId: catch_all__full_path__get
+      parameters:
+      - in: path
+        name: full_path
+        required: true
+        schema:
+          title: Full Path
+          type: string
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+        '422':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+          description: Validation Error
+      summary: Catch All
   /healthz:
     get:
       description: Health check endpoint. Expecting an empty response with status
@@ -1076,39 +1109,6 @@ paths:
           description: Successful Response
       tags:
       - Infrastructure
-  /ui/:
-    get:
-      operationId: serve_chat_html__get
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-      summary: Serve Chat Html
-  /ui/{full_path}:
-    get:
-      operationId: catch_all__full_path__get
-      parameters:
-      - in: path
-        name: full_path
-        required: true
-        schema:
-          title: Full Path
-          type: string
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-        '422':
-          content:
-            application/json:
-              schema:
-                $ref: '#/components/schemas/HTTPValidationError'
-          description: Validation Error
-      summary: Catch All
   /v1/chat/completions:
     post:
       operationId: create_chat_completion_chat_completions_post

bentoml/bentos/gemma/2b-instruct-fp16-da36/apis/openapi.yaml renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/apis/openapi.yaml

@@ -1,8 +1,8 @@
 service: service:VLLM
 name: gemma
-version: 2b-instruct-fp16-da36
+version: 2b-instruct-fp16-26b3
 bentoml_version: 1.2.19
-creation_time: '2024-07-05T07:31:05.934035+00:00'
+creation_time: '2024-07-08T07:57:26.100524+00:00'
 labels:
   openllm_alias: 2b,2b-instruct
   openllm_hf_model_id: google/gemma-2b-it

bentoml/bentos/gemma/2b-instruct-fp16-da36/apis/schema.json renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/apis/schema.json

@@ -143,5 +143,5 @@ yarl==1.9.4
 zipp==3.19.2
 
 # The following packages are considered to be unsafe in a requirements file:
-pip==24.1.1
+pip==24.1.2
 setuptools==70.2.0

bentoml/bentos/gemma/2b-instruct-fp16-da36/bento.yaml renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/bento.yaml

@@ -85,7 +85,7 @@ async def catch_all(full_path: str):
 
 
 @bentoml.mount_asgi_app(openai_api_app, path="/v1")
-@bentoml.mount_asgi_app(ui_app, path="/ui")
+@bentoml.mount_asgi_app(ui_app, path="/chat")
 @bentoml.service(**SERVICE_CONFIG)
 class VLLM:
     def __init__(self) -> None:

bentoml/bentos/gemma/2b-instruct-fp16-da36/env/docker/Dockerfile renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/docker/Dockerfile

@@ -1036,6 +1036,39 @@ paths:
       tags:
       - Service APIs
       x-bentoml-name: generate
+  /chat/:
+    get:
+      operationId: serve_chat_html__get
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+      summary: Serve Chat Html
+  /chat/{full_path}:
+    get:
+      operationId: catch_all__full_path__get
+      parameters:
+      - in: path
+        name: full_path
+        required: true
+        schema:
+          title: Full Path
+          type: string
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+        '422':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+          description: Validation Error
+      summary: Catch All
   /healthz:
     get:
       description: Health check endpoint. Expecting an empty response with status
@@ -1076,39 +1109,6 @@ paths:
           description: Successful Response
       tags:
       - Infrastructure
-  /ui/:
-    get:
-      operationId: serve_chat_html__get
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-      summary: Serve Chat Html
-  /ui/{full_path}:
-    get:
-      operationId: catch_all__full_path__get
-      parameters:
-      - in: path
-        name: full_path
-        required: true
-        schema:
-          title: Full Path
-          type: string
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-        '422':
-          content:
-            application/json:
-              schema:
-                $ref: '#/components/schemas/HTTPValidationError'
-          description: Validation Error
-      summary: Catch All
   /v1/chat/completions:
     post:
       operationId: create_chat_completion_chat_completions_post

bentoml/bentos/gemma/2b-instruct-fp16-da36/env/docker/entrypoint.sh renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/docker/entrypoint.sh

@@ -1,8 +1,8 @@
 service: service:VLLM
 name: gemma
-version: 7b-instruct-awq-4bit-6a74
+version: 7b-instruct-awq-4bit-d670
 bentoml_version: 1.2.19
-creation_time: '2024-07-05T07:32:14.773611+00:00'
+creation_time: '2024-07-08T07:59:18.619602+00:00'
 labels:
   openllm_alias: 7b-4bit,7b-instruct-4bit
   openllm_hf_model_id: casperhansen/gemma-7b-it-awq

bentoml/bentos/gemma/2b-instruct-fp16-da36/env/python/install.sh renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/python/install.sh

@@ -143,5 +143,5 @@ yarl==1.9.4
 zipp==3.19.2
 
 # The following packages are considered to be unsafe in a requirements file:
-pip==24.1.1
+pip==24.1.2
 setuptools==70.2.0

bentoml/bentos/gemma/7b-instruct-awq-4bit-6a74/env/python/requirements.lock.txt renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/python/requirements.lock.txt

@@ -85,7 +85,7 @@ async def catch_all(full_path: str):
 
 
 @bentoml.mount_asgi_app(openai_api_app, path="/v1")
-@bentoml.mount_asgi_app(ui_app, path="/ui")
+@bentoml.mount_asgi_app(ui_app, path="/chat")
 @bentoml.service(**SERVICE_CONFIG)
 class VLLM:
     def __init__(self) -> None:

bentoml/bentos/gemma/2b-instruct-fp16-da36/env/python/requirements.txt renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/python/requirements.txt

@@ -1036,6 +1036,39 @@ paths:
       tags:
       - Service APIs
       x-bentoml-name: generate
+  /chat/:
+    get:
+      operationId: serve_chat_html__get
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+      summary: Serve Chat Html
+  /chat/{full_path}:
+    get:
+      operationId: catch_all__full_path__get
+      parameters:
+      - in: path
+        name: full_path
+        required: true
+        schema:
+          title: Full Path
+          type: string
+      responses:
+        '200':
+          content:
+            application/json:
+              schema: {}
+          description: Successful Response
+        '422':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+          description: Validation Error
+      summary: Catch All
   /healthz:
     get:
       description: Health check endpoint. Expecting an empty response with status
@@ -1076,39 +1109,6 @@ paths:
           description: Successful Response
       tags:
       - Infrastructure
-  /ui/:
-    get:
-      operationId: serve_chat_html__get
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-      summary: Serve Chat Html
-  /ui/{full_path}:
-    get:
-      operationId: catch_all__full_path__get
-      parameters:
-      - in: path
-        name: full_path
-        required: true
-        schema:
-          title: Full Path
-          type: string
-      responses:
-        '200':
-          content:
-            application/json:
-              schema: {}
-          description: Successful Response
-        '422':
-          content:
-            application/json:
-              schema:
-                $ref: '#/components/schemas/HTTPValidationError'
-          description: Validation Error
-      summary: Catch All
   /v1/chat/completions:
     post:
       operationId: create_chat_completion_chat_completions_post

bentoml/bentos/gemma/2b-instruct-fp16-da36/env/python/version.txt renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/env/python/version.txt

@@ -1,8 +1,8 @@
 service: service:VLLM
 name: gemma
-version: 7b-instruct-fp16-dafc
+version: 7b-instruct-fp16-ae3e
 bentoml_version: 1.2.19
-creation_time: '2024-07-05T07:31:38.255529+00:00'
+creation_time: '2024-07-08T07:58:22.853090+00:00'
 labels:
   openllm_alias: 7b,7b-instruct
   openllm_hf_model_id: google/gemma-7b-it

bentoml/bentos/gemma/2b-instruct-fp16-da36/src/bento_constants.py renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/src/bento_constants.py

@@ -143,5 +143,5 @@ yarl==1.9.4
 zipp==3.19.2
 
 # The following packages are considered to be unsafe in a requirements file:
-pip==24.1.1
+pip==24.1.2
 setuptools==70.2.0

bentoml/bentos/gemma/2b-instruct-fp16-da36/src/bentofile.yaml renamed to bentoml/bentos/gemma/2b-instruct-fp16-26b3/src/bentofile.yaml

@@ -85,7 +85,7 @@ async def catch_all(full_path: str):
 
 
 @bentoml.mount_asgi_app(openai_api_app, path="/v1")
-@bentoml.mount_asgi_app(ui_app, path="/ui")
+@bentoml.mount_asgi_app(ui_app, path="/chat")
 @bentoml.service(**SERVICE_CONFIG)
 class VLLM:
     def __init__(self) -> None: