refactor: consolidate DeepSeek model entries and add fallback groups

litellm/config.yaml

@@ -26,23 +26,12 @@ model_list:
   # ═══════════════════════════════════════════════
 
   # DeepSeek V3.2 via DeepInfra ($0.26 in / $0.38 out per M)
-  - model_name: deepseek-v3
-    litellm_params:
-      model: deepinfra/deepseek-ai/DeepSeek-V3.2
-      api_key: os.environ/DEEPINFRA_API_KEY
-
   - model_name: deepseek-v3.2
     litellm_params:
       model: deepinfra/deepseek-ai/DeepSeek-V3.2
       api_key: os.environ/DEEPINFRA_API_KEY
 
   # DeepSeek V3.2 fallback via SiliconFlow ($0.27 in / $0.42 out per M)
-  - model_name: deepseek-v3
-    litellm_params:
-      model: openai/deepseek-ai/DeepSeek-V3.2
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
   - model_name: deepseek-v3.2
     litellm_params:
       model: openai/deepseek-ai/DeepSeek-V3.2
@@ -207,8 +196,8 @@ model_list:
       model: openrouter/openai/gpt-4.1
       api_key: os.environ/OPENROUTER_API_KEY
 
-  # DeepSeek V3 last-resort fallback via OpenRouter
-  - model_name: deepseek-v3
+  # DeepSeek V3.2 last-resort fallback via OpenRouter
+  - model_name: deepseek-v3.2
     litellm_params:
       model: openrouter/deepseek/deepseek-chat-v3-0324
       api_key: os.environ/OPENROUTER_API_KEY
@@ -216,6 +205,15 @@ model_list:
 general_settings:
   master_key: os.environ/LITELLM_MASTER_KEY
 
+  # ── Model group fallbacks (when model misbehaves mid-stream) ──
+  fallbacks:
+    - deepseek-v3.2: [gpt-oss, kimi-k2]
+    - gpt-oss: [deepseek-v3.2, nemotron-super]
+    - kimi-k2: [deepseek-v3.2, gpt-oss]
+    - kimi-k2.5: [deepseek-v3.2, gpt-oss]
+    - glm-4.7: [deepseek-v3.2, gpt-oss]
+    - glm-5: [deepseek-v3.2, gpt-oss]
+
 litellm_settings:
   drop_params: true
   set_verbose: false