Support for prem_1b_chat

prem_utils/connectors/prem.py

@@ -11,6 +11,8 @@
 from prem_utils.connectors import utils as connector_utils
 from prem_utils.connectors.base import BaseConnector
 
+DEFAULT_MAX_NEW_TOKENS = 1024
+
 
 class PremConnector(BaseConnector):
     def __init__(
@@ -45,7 +47,7 @@ async def chat_completion(
         self,
         model: str,
         messages: list[dict[str, str]],
-        max_tokens: int | None = 128,
+        max_tokens: int | None = DEFAULT_MAX_NEW_TOKENS,
         stream: bool = False,
         temperature: float = 1,
         top_p: float = 0.99,

prem_utils/models.json

@@ -1452,10 +1452,11 @@
                     ]
                 },
                 {
+
                     "tier": "low",
                     "slug": "prem-1b-chat",
                     "model_type": "text2text",
-                    "coming_soon": true,
+                    "context_window": 8192,
                     "group": "slm",
                     "environment": [
                         "development"