Enable Apriel 1.6, and add an announcement banner

app.py

@@ -30,13 +30,17 @@ BUTTON_WIDTH = 160
 DEFAULT_OPT_OUT_VALUE = DEBUG_MODE
 
 # If DEBUG_MODEL is True, use an alternative model (without reasoning) for testing
-DEFAULT_MODEL_NAME = "Apriel-1.5-15B-thinker" if not DEBUG_MODEL else "Apriel-1.5-15B-thinker"
-# DEFAULT_MODEL_NAME = "Apriel-1.6-15B-Thinker" if not DEBUG_MODEL else "Apriel-1.6-15B-Thinker"
+# DEFAULT_MODEL_NAME = "Apriel-1.5-15B-thinker" if not DEBUG_MODEL else "Apriel-1.5-15B-thinker"
+DEFAULT_MODEL_NAME = "Apriel-1.6-15B-Thinker"
 
-SHOW_BANNER = False
-BANNER_MARKDOWN = """
+SHOW_BANNER = True
+INFO_BANNER_MARKDOWN = """
                     <span class="banner-message-text">ℹ️ This app has been updated to use the recommended temperature of 0.6. We had set it to 0.8 earlier and expect 0.6 to be better. Please provide feedback using the model link.</span>
                     """
+NEW_MODEL_BANNER_MARKDOWN = """
+                    <span class="banner-message-text"><span class="banner-message-emoji">🚀</span> Now running [Apriel-1.6-15B-Thinker](https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker) - 30% more efficient, frontier-class reasoning</span>
+                    """
+BANNER_MARKDOWN = NEW_MODEL_BANNER_MARKDOWN
 
 BUTTON_ENABLED = gr.update(interactive=True)
 BUTTON_DISABLED = gr.update(interactive=False)

styles.css

@@ -1,9 +1,11 @@
 :root {
     --color-grey-50: #f9fafb;
     --banner-background: var(--secondary-400);
-    --banner-text-color: var(--primary-100);
+    --banner-text-color: var(--primary-200);
     --banner-background-dark: var(--secondary-800);
     --banner-text-color-dark: var(--primary-100);
+    --banner-text-a-color: var(--primary-100);
+    --banner-text-a-color-dark: var(--secondary-100);
     --banner-chrome-height: calc(16px + 43px);
 
     /* Chrome height with no banner */
@@ -28,11 +30,20 @@
 }
 
 .banner-message-text {
-    font-size: 13px;
+    font-size: 14px;
     font-weight: bolder;
     color: var(--banner-text-color) !important;
 }
 
+.banner-message-emoji {
+    margin-right: 3px;
+    font-size: 16px;
+}
+
+.banner-message-text a {
+    color: var(--banner-text-a-color) !important;
+}
+
 body.dark .banner-message {
     background-color: var(--banner-background-dark) !important;
 }
@@ -40,6 +51,10 @@ body.dark .gradio-container .contain .banner-message .banner-message-text {
     color: var(--banner-text-color-dark) !important;
 }
 
+body.dark .gradio-container .contain .banner-message .banner-message-text a {
+    color: var(--banner-text-a-color-dark) !important;
+}
+
 .toast-body {
     background-color: var(--color-grey-50);
 }

utils.py

@@ -12,34 +12,34 @@ DEBUG_MODE = False or os.environ.get("DEBUG_MODE") == "True"
 DEBUG_MODEL = False or os.environ.get("DEBUG_MODEL") == "True"
 
 models_config = {
-    # "Apriel-1.6-15B-Thinker": {
-    #     "MODEL_DISPLAY_NAME": "Apriel-1.6-15B-Thinker",
-    #     "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker",
-    #     "MODEL_NAME": os.environ.get("MODEL_NAME_APRIEL_1_6_15B"),
-    #     "VLLM_API_URL": os.environ.get("VLLM_API_URL_APRIEL_1_6_15B"),
-    #     "VLLM_API_URL_LIST": os.environ.get("VLLM_API_URL_LIST_APRIEL_1_6_15B"),
-    #     "AUTH_TOKEN": os.environ.get("AUTH_TOKEN"),
-    #     "REASONING": True,
-    #     "MULTIMODAL": True,
-    #     "TEMPERATURE": 0.6,
-    #     "OUTPUT_TAG_START": "[BEGIN FINAL RESPONSE]",
-    #     "OUTPUT_TAG_END": "",
-    #     "OUTPUT_STOP_TOKEN": "<|end|>"
-    # },
-    "Apriel-1.5-15B-thinker": {
-        "MODEL_DISPLAY_NAME": "Apriel-1.5-15B-thinker",
-        "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.5-15b-Thinker",
-        "MODEL_NAME": os.environ.get("MODEL_NAME_APRIEL_1_5_15B"),
-        "VLLM_API_URL": os.environ.get("VLLM_API_URL_APRIEL_1_5_15B"),
-        "VLLM_API_URL_LIST": os.environ.get("VLLM_API_URL_LIST_APRIEL_1_5_15B"),
+    "Apriel-1.6-15B-Thinker": {
+        "MODEL_DISPLAY_NAME": "Apriel-1.6-15B-Thinker",
+        "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker",
+        "MODEL_NAME": os.environ.get("MODEL_NAME_APRIEL_1_6_15B"),
+        "VLLM_API_URL": os.environ.get("VLLM_API_URL_APRIEL_1_6_15B"),
+        "VLLM_API_URL_LIST": os.environ.get("VLLM_API_URL_LIST_APRIEL_1_6_15B"),
         "AUTH_TOKEN": os.environ.get("AUTH_TOKEN"),
         "REASONING": True,
         "MULTIMODAL": True,
         "TEMPERATURE": 0.6,
         "OUTPUT_TAG_START": "[BEGIN FINAL RESPONSE]",
-        "OUTPUT_TAG_END": "[END FINAL RESPONSE]",
+        "OUTPUT_TAG_END": "",
         "OUTPUT_STOP_TOKEN": "<|end|>"
     },
+    # "Apriel-1.5-15B-thinker": {
+    #     "MODEL_DISPLAY_NAME": "Apriel-1.5-15B-thinker",
+    #     "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-1.5-15b-Thinker",
+    #     "MODEL_NAME": os.environ.get("MODEL_NAME_APRIEL_1_5_15B"),
+    #     "VLLM_API_URL": os.environ.get("VLLM_API_URL_APRIEL_1_5_15B"),
+    #     "VLLM_API_URL_LIST": os.environ.get("VLLM_API_URL_LIST_APRIEL_1_5_15B"),
+    #     "AUTH_TOKEN": os.environ.get("AUTH_TOKEN"),
+    #     "REASONING": True,
+    #     "MULTIMODAL": True,
+    #     "TEMPERATURE": 0.6,
+    #     "OUTPUT_TAG_START": "[BEGIN FINAL RESPONSE]",
+    #     "OUTPUT_TAG_END": "[END FINAL RESPONSE]",
+    #     "OUTPUT_STOP_TOKEN": "<|end|>"
+    # },
     # "Apriel-Nemotron-15b-Thinker": {
     #     "MODEL_DISPLAY_NAME": "Apriel-Nemotron-15b-Thinker",
     #     "MODEL_HF_URL": "https://huggingface.co/ServiceNow-AI/Apriel-Nemotron-15b-Thinker",