From 69b70121efff8d6fe46ffe56f5ae80c48577555a Mon Sep 17 00:00:00 2001 From: Ishaan Jaff Date: Wed, 26 Jun 2024 11:26:14 -0700 Subject: [PATCH] add chat-bison-32k@002 --- ...odel_prices_and_context_window_backup.json | 30 +++++++++++++++++++ model_prices_and_context_window.json | 30 +++++++++++++++++++ 2 files changed, 60 insertions(+) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index f9453bc0f94..20f5ecec975 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -1147,6 +1147,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1157,6 +1159,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1167,6 +1171,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1177,6 +1183,20 @@ "max_output_tokens": 8192, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, + "litellm_provider": "vertex_ai-chat-models", + "mode": "chat", + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + }, + "chat-bison-32k@002": { + "max_tokens": 8192, + "max_input_tokens": 32000, + "max_output_tokens": 8192, + "input_cost_per_token": 0.000000125, + "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1187,6 +1207,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-text-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1197,6 +1219,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1237,6 +1261,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1247,6 +1273,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1257,6 +1285,8 @@ "max_output_tokens": 8192, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index f9453bc0f94..20f5ecec975 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -1147,6 +1147,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1157,6 +1159,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1167,6 +1171,8 @@ "max_output_tokens": 4096, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1177,6 +1183,20 @@ "max_output_tokens": 8192, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, + "litellm_provider": "vertex_ai-chat-models", + "mode": "chat", + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + }, + "chat-bison-32k@002": { + "max_tokens": 8192, + "max_input_tokens": 32000, + "max_output_tokens": 8192, + "input_cost_per_token": 0.000000125, + "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1187,6 +1207,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-text-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1197,6 +1219,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1237,6 +1261,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1247,6 +1273,8 @@ "max_output_tokens": 1024, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -1257,6 +1285,8 @@ "max_output_tokens": 8192, "input_cost_per_token": 0.000000125, "output_cost_per_token": 0.000000125, + "input_cost_per_character": 0.00000025, + "output_cost_per_character": 0.0000005, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models"