2023-07-05 21:21:08 +00:00
|
|
|
MAX_TOKENS = {
|
2023-09-05 05:40:05 +00:00
|
|
|
'text-embedding-ada-002': 8000,
|
2024-04-23 00:25:32 +00:00
|
|
|
'gpt-3.5-turbo': 16000,
|
|
|
|
|
'gpt-3.5-turbo-0125': 16000,
|
2023-07-05 21:21:08 +00:00
|
|
|
'gpt-3.5-turbo-0613': 4000,
|
2024-04-23 00:25:32 +00:00
|
|
|
'gpt-3.5-turbo-1106': 16000,
|
2023-07-05 21:21:08 +00:00
|
|
|
'gpt-3.5-turbo-16k': 16000,
|
|
|
|
|
'gpt-3.5-turbo-16k-0613': 16000,
|
|
|
|
|
'gpt-4': 8000,
|
|
|
|
|
'gpt-4-0613': 8000,
|
|
|
|
|
'gpt-4-32k': 32000,
|
2024-06-03 15:58:31 +00:00
|
|
|
'gpt-4-1106-preview': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-02-01 07:46:04 +00:00
|
|
|
'gpt-4-0125-preview': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-05-14 01:24:34 +00:00
|
|
|
'gpt-4o': 128000, # 128K, but may be limited by config.max_model_tokens
|
|
|
|
|
'gpt-4o-2024-05-13': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-03-26 05:47:05 +00:00
|
|
|
'gpt-4-turbo-preview': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-04-10 05:44:38 +00:00
|
|
|
'gpt-4-turbo-2024-04-09': 128000, # 128K, but may be limited by config.max_model_tokens
|
|
|
|
|
'gpt-4-turbo': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-07-23 00:41:04 +00:00
|
|
|
'gpt-4o-mini': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-07-23 00:43:47 +00:00
|
|
|
'gpt-4o-mini-2024-07-18': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-08-07 00:28:51 +00:00
|
|
|
'gpt-4o-2024-08-06': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-11-21 15:10:27 +00:00
|
|
|
'gpt-4o-2024-11-20': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-09-13 01:22:09 +00:00
|
|
|
'o1-mini': 128000, # 128K, but may be limited by config.max_model_tokens
|
|
|
|
|
'o1-mini-2024-09-12': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-09-13 01:23:18 +00:00
|
|
|
'o1-preview': 128000, # 128K, but may be limited by config.max_model_tokens
|
|
|
|
|
'o1-preview-2024-09-12': 128000, # 128K, but may be limited by config.max_model_tokens
|
2024-12-19 16:11:45 +00:00
|
|
|
'o1-2024-12-17': 204800, # 200K, but may be limited by config.max_model_tokens
|
2024-12-22 03:33:26 +00:00
|
|
|
'o1': 204800, # 200K, but may be limited by config.max_model_tokens
|
2025-02-02 11:02:53 +00:00
|
|
|
'o3-mini': 204800, # 200K, but may be limited by config.max_model_tokens
|
|
|
|
|
'o3-mini-2025-01-31': 204800, # 200K, but may be limited by config.max_model_tokens
|
2023-08-03 23:05:46 +00:00
|
|
|
'claude-instant-1': 100000,
|
|
|
|
|
'claude-2': 100000,
|
|
|
|
|
'command-nightly': 4096,
|
2025-01-17 04:42:36 +00:00
|
|
|
'deepseek/deepseek-chat': 128000, # 128K, but may be limited by config.max_model_tokens
|
2025-01-22 13:32:18 +00:00
|
|
|
'deepseek/deepseek-reasoner': 64000, # 64K, but may be limited by config.max_model_tokens
|
2023-08-03 23:05:46 +00:00
|
|
|
'replicate/llama-2-70b-chat:2c1608e18606fad2812020dc541930f2d0495ce32eee50074220b87300bc16e1': 4096,
|
2023-11-07 09:13:08 +00:00
|
|
|
'meta-llama/Llama-2-7b-chat-hf': 4096,
|
|
|
|
|
'vertex_ai/codechat-bison': 6144,
|
|
|
|
|
'vertex_ai/codechat-bison-32k': 32000,
|
2024-06-13 05:34:14 +00:00
|
|
|
'vertex_ai/claude-3-haiku@20240307': 100000,
|
2024-11-06 00:52:58 +00:00
|
|
|
'vertex_ai/claude-3-5-haiku@20241022': 100000,
|
2024-06-13 05:34:14 +00:00
|
|
|
'vertex_ai/claude-3-sonnet@20240229': 100000,
|
|
|
|
|
'vertex_ai/claude-3-opus@20240229': 100000,
|
2024-06-25 08:32:17 +00:00
|
|
|
'vertex_ai/claude-3-5-sonnet@20240620': 100000,
|
2024-10-23 00:11:33 +00:00
|
|
|
'vertex_ai/claude-3-5-sonnet-v2@20241022': 100000,
|
2024-06-24 12:01:33 +00:00
|
|
|
'vertex_ai/gemini-1.5-pro': 1048576,
|
2024-07-09 12:29:50 +00:00
|
|
|
'vertex_ai/gemini-1.5-flash': 1048576,
|
2025-01-20 09:29:37 +00:00
|
|
|
'vertex_ai/gemini-2.0-flash-exp': 1048576,
|
2024-07-09 12:29:50 +00:00
|
|
|
'vertex_ai/gemma2': 8200,
|
2024-10-28 23:00:16 +00:00
|
|
|
'gemini/gemini-1.5-pro': 1048576,
|
|
|
|
|
'gemini/gemini-1.5-flash': 1048576,
|
2024-12-16 16:37:19 +00:00
|
|
|
'gemini/gemini-2.0-flash-exp': 1048576,
|
2023-11-07 09:13:08 +00:00
|
|
|
'codechat-bison': 6144,
|
|
|
|
|
'codechat-bison-32k': 32000,
|
2023-11-28 11:11:40 +00:00
|
|
|
'anthropic.claude-instant-v1': 100000,
|
|
|
|
|
'anthropic.claude-v1': 100000,
|
2024-03-06 06:20:08 +00:00
|
|
|
'anthropic.claude-v2': 100000,
|
|
|
|
|
'anthropic/claude-3-opus-20240229': 100000,
|
2024-06-21 08:30:52 +00:00
|
|
|
'anthropic/claude-3-5-sonnet-20240620': 100000,
|
2024-10-23 00:10:22 +00:00
|
|
|
'anthropic/claude-3-5-sonnet-20241022': 100000,
|
2024-11-06 00:52:58 +00:00
|
|
|
'anthropic/claude-3-5-haiku-20241022': 100000,
|
2024-03-13 02:14:04 +00:00
|
|
|
'bedrock/anthropic.claude-instant-v1': 100000,
|
|
|
|
|
'bedrock/anthropic.claude-v2': 100000,
|
|
|
|
|
'bedrock/anthropic.claude-v2:1': 100000,
|
|
|
|
|
'bedrock/anthropic.claude-3-sonnet-20240229-v1:0': 100000,
|
2024-03-14 07:58:44 +00:00
|
|
|
'bedrock/anthropic.claude-3-haiku-20240307-v1:0': 100000,
|
2024-11-07 02:12:52 +00:00
|
|
|
'bedrock/anthropic.claude-3-5-haiku-20241022-v1:0': 100000,
|
2024-07-04 09:23:36 +00:00
|
|
|
'bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0': 100000,
|
2024-10-23 01:08:04 +00:00
|
|
|
'bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0': 100000,
|
2024-12-23 15:06:21 +00:00
|
|
|
"bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0": 100000,
|
2024-08-11 06:21:34 +00:00
|
|
|
'claude-3-5-sonnet': 100000,
|
2024-04-21 06:21:45 +00:00
|
|
|
'groq/llama3-8b-8192': 8192,
|
|
|
|
|
'groq/llama3-70b-8192': 8192,
|
2024-12-27 16:51:33 +00:00
|
|
|
'groq/llama-3.1-8b-instant': 8192,
|
2024-12-30 02:35:02 +00:00
|
|
|
'groq/llama-3.3-70b-versatile': 128000,
|
2024-07-27 17:28:42 +00:00
|
|
|
'groq/mixtral-8x7b-32768': 32768,
|
2024-12-27 16:51:33 +00:00
|
|
|
'groq/gemma2-9b-it': 8192,
|
2024-06-04 07:09:30 +00:00
|
|
|
'ollama/llama3': 4096,
|
2024-07-29 05:32:34 +00:00
|
|
|
'watsonx/meta-llama/llama-3-8b-instruct': 4096,
|
|
|
|
|
"watsonx/meta-llama/llama-3-70b-instruct": 4096,
|
|
|
|
|
"watsonx/meta-llama/llama-3-405b-instruct": 16384,
|
|
|
|
|
"watsonx/ibm/granite-13b-chat-v2": 8191,
|
|
|
|
|
"watsonx/ibm/granite-34b-code-instruct": 8191,
|
|
|
|
|
"watsonx/mistralai/mistral-large": 32768,
|
2023-07-05 21:21:08 +00:00
|
|
|
}
|
2025-02-02 11:01:16 +00:00
|
|
|
|
|
|
|
|
USER_MESSAGE_ONLY_MODELS = [
|
|
|
|
|
"deepseek/deepseek-reasoner",
|
|
|
|
|
"o1-mini",
|
2025-02-16 08:18:52 +00:00
|
|
|
"o1-mini-2024-09-12",
|
|
|
|
|
"o1-preview"
|
2025-02-02 11:01:16 +00:00
|
|
|
]
|
2025-02-16 08:17:59 +00:00
|
|
|
|
|
|
|
|
NO_SUPPORT_TEMPERATURE_MODELS = [
|
|
|
|
|
"deepseek/deepseek-reasoner",
|
|
|
|
|
"o1-mini",
|
|
|
|
|
"o1-mini-2024-09-12",
|
|
|
|
|
"o1",
|
|
|
|
|
"o1-2024-12-17",
|
|
|
|
|
"o3-mini",
|
|
|
|
|
"o3-mini-2025-01-31",
|
2025-02-16 08:18:52 +00:00
|
|
|
"o1-preview"
|
2025-02-16 08:17:59 +00:00
|
|
|
]
|