UPDATE: change default model load workflow

app.py

@@ -87,11 +87,19 @@ for model_config in CONFIG.MODELS:
     )
     tmp_pipeline = PIPELINE(tmp_model, model_config.VOCAB)
 
-    if model_config.DEFAULT:
-        if model_config.REASONING:
-            DEFAULT_REASONING_MODEL_NAME = model_config.SERVICE_NAME
-        else:
-            DEFALUT_MODEL_NAME = model_config.SERVICE_NAME
+    if model_config.DEFAULT_CHAT:
+        if DEFALUT_MODEL_NAME != None:
+            logger.info(
+                f"Load Model - Replace `DEFALUT_MODEL_NAME` from `{DEFALUT_MODEL_NAME}` to `{model_config.SERVICE_NAME}`"
+            )
+        DEFALUT_MODEL_NAME = model_config.SERVICE_NAME
+
+    if model_config.DEFAULT_REASONING:
+        if DEFAULT_REASONING_MODEL_NAME != None:
+            logger.info(
+                f"Load Model - Replace `DEFAULT_REASONING_MODEL_NAME` from `{DEFAULT_REASONING_MODEL_NAME}` to `{model_config.SERVICE_NAME}`"
+            )
+        DEFAULT_REASONING_MODEL_NAME = model_config.SERVICE_NAME
 
     print(model_config.DEFAULT_SAMPLER)
 
@@ -101,8 +109,10 @@ for model_config in CONFIG.MODELS:
     MODEL_STORAGE[model_config.SERVICE_NAME].pipeline = tmp_pipeline
 
 
-logger.info(f"DEFALUT_MODEL_NAME is `{DEFALUT_MODEL_NAME}`")
-logger.info(f"DEFAULT_REASONING_MODEL_NAME is `{DEFAULT_REASONING_MODEL_NAME}`")
+logger.info(f"Load Model - DEFALUT_MODEL_NAME is `{DEFALUT_MODEL_NAME}`")
+logger.info(
+    f"Load Model - DEFAULT_REASONING_MODEL_NAME is `{DEFAULT_REASONING_MODEL_NAME}`"
+)
 
 
 class ChatCompletionRequest(BaseModel):

config.production.yaml

@@ -1,16 +1,17 @@
 HOST: "0.0.0.0"
 PORT: 7860
 STRATEGY: "cuda fp16"
-RWKV_CUDA_ON: False
+RWKV_CUDA_ON: True
 CHUNK_LEN: 256
 MODELS:
-  - SERVICE_NAME: "RWKV-x070-World-0.1B-v2.8-20241210-ctx4096"
-    DOWNLOAD_MODEL_FILE_NAME: "RWKV-x070-World-0.1B-v2.8-20241210-ctx4096.pth"
-    DOWNLOAD_MODEL_REPO_ID: "BlinkDL/rwkv-7-world"
-    REASONING: False
-    DEFAULT: True
+  - SERVICE_NAME: "rwkv7-g1-0.1b-20250307-ctx4096"
+    DOWNLOAD_MODEL_FILE_NAME: "rwkv7-g1-0.1b-20250307-ctx4096.pth"
+    DOWNLOAD_MODEL_REPO_ID: "BlinkDL/rwkv7-g1"
+    REASONING: True
+    DEFAULT_CHAT: True
+    DEFAULT_REASONING: True
     DEFAULT_SAMPLER:
-      max_tokens: 512
+      max_tokens: 4096
       temperature: 1.0
       top_p: 0.3
       presence_penalty: 0.5
@@ -20,11 +21,10 @@ MODELS:
         - "\n\n"
       stop_tokens:
         - 0
-  - SERVICE_NAME: "rwkv7-g1-0.1b-20250307-ctx4096"
-    DOWNLOAD_MODEL_FILE_NAME: "rwkv7-g1-0.1b-20250307-ctx4096.pth"
-    DOWNLOAD_MODEL_REPO_ID: "BlinkDL/rwkv7-g1"
+  - SERVICE_NAME: "RWKV7-G1-1.5B-16%trained-20250308-ctx4k"
+    DOWNLOAD_MODEL_FILE_NAME: "RWKV7-G1-1.5B-16%trained-20250308-ctx4k.pth"
+    DOWNLOAD_MODEL_REPO_ID: "BlinkDL/rwkv-7-world"
     REASONING: True
-    DEFAULT: True
     DEFAULT_SAMPLER:
       max_tokens: 4096
       temperature: 1.0

config.py

@@ -49,7 +49,8 @@ class ModelConfig(BaseModel):
         False, description="Whether reasoning is enabled for this model."
     )
 
-    DEFAULT: bool = Field(False, description="Whether this model is the default model.")
+    DEFAULT_CHAT: bool = Field(False, description="Whether this model is the default chat model.")
+    DEFAULT_REASONING: bool = Field(False, description="Whether this model is the default reasoning model.")
     DEFAULT_SAMPLER: SamplerConfig = Field(
         SamplerConfig(), description="Default sampler configuration for this model."
     )