petrov826 commited on
Commit
95fed74
·
verified ·
1 Parent(s): 48a69ce

mix pro and flash model in model_list

Browse files
Files changed (1) hide show
  1. app.py +29 -20
app.py CHANGED
@@ -46,34 +46,43 @@ rate_limiter = RateLimiter()
46
  class BasicAgent:
47
  def __init__(self):
48
  model = LiteLLMRouterModel(
49
- model_id="gemini-pro-family",
50
  model_list=[
51
  {
52
- "model_name": "gemini-pro-family",
53
  "litellm_params": {
54
  "model": "gemini/gemini-2.5-pro",
55
  "rpm": 5,
56
  "tpm": 250_000,
57
  },
58
- # "fallbacks": ["gemini/gemini-2.0-flash"],
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
59
  },
60
- # {
61
- # "model_name": "gemini-flash-family",
62
- # "litellm_params": {
63
- # "model": "gemini/gemini-2.0-flash",
64
- # "rpm": 15,
65
- # "tpm": 1_000_000,
66
- # },
67
- # "fallbacks": ["gemini/gemini-2.0-flash-lite"]
68
- # },
69
- # {
70
- # "model_name": "gemini-flash-family",
71
- # "litellm_params": {
72
- # "model": "gemini/gemini-2.0-flash-lite",
73
- # "rpm": 30,
74
- # "tpm": 1_000_000,
75
- # },
76
- # },
77
  ],
78
  client_kwargs={
79
  "routing_strategy": "usage-based-routing-v2"
 
46
  class BasicAgent:
47
  def __init__(self):
48
  model = LiteLLMRouterModel(
49
+ model_id="gemini-family",
50
  model_list=[
51
  {
52
+ "model_name": "gemini-family",
53
  "litellm_params": {
54
  "model": "gemini/gemini-2.5-pro",
55
  "rpm": 5,
56
  "tpm": 250_000,
57
  },
58
+ "fallbacks": ["gemini/gemini-2.5-flash"],
59
+ },
60
+ {
61
+ "model_name": "gemini-family",
62
+ "litellm_params": {
63
+ "model": "gemini/gemini-2.5-flash",
64
+ "rpm": 10,
65
+ "tpm": 250_000,
66
+ },
67
+ "fallbacks": ["gemini/gemini-2.0-flash"]
68
+ },
69
+ {
70
+ "model_name": "gemini-family",
71
+ "litellm_params": {
72
+ "model": "gemini/gemini-2.0-flash",
73
+ "rpm": 15,
74
+ "tpm": 1_000_000,
75
+ },
76
+ "fallbacks": ["gemini/gemini-2.0-flash-lite"]
77
+ },
78
+ {
79
+ "model_name": "gemini-family",
80
+ "litellm_params": {
81
+ "model": "gemini/gemini-2.0-flash-lite",
82
+ "rpm": 30,
83
+ "tpm": 1_000_000,
84
+ },
85
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86
  ],
87
  client_kwargs={
88
  "routing_strategy": "usage-based-routing-v2"