diff options
| author | sigoden <sigoden@gmail.com> | 2024-07-27 18:29:56 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-07-27 18:29:56 +0800 |
| commit | adf6716c8436b53854eb55588d258f4326580a4d (patch) | |
| tree | e30661b2725d61b097faa51d472b3bff73b0c457 /models.yaml | |
| parent | 2eed63f014deae539b22360c44c40e9ab2fef9a0 (diff) | |
| download | aichat-adf6716c8436b53854eb55588d258f4326580a4d.tar.gz | |
feat: bedrock client switch to converse api and support cohere models (#747)
Diffstat (limited to 'models.yaml')
| -rw-r--r-- | models.yaml | 59 |
1 files changed, 36 insertions, 23 deletions
diff --git a/models.yaml b/models.yaml index a2bdfae..fc78583 100644 --- a/models.yaml +++ b/models.yaml @@ -368,8 +368,9 @@ # docs: # - https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html#model-ids-arns # - https://aws.amazon.com/bedrock/pricing/ + # - https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference.html # notes: - # - get max_output_tokens info from playground + # - except for Claude, other models do not support streaming function calling models: - name: anthropic.claude-3-5-sonnet-20240620-v1:0 max_input_tokens: 200000 @@ -405,44 +406,56 @@ supports_function_calling: true - name: meta.llama3-1-405b-instruct-v1:0 max_input_tokens: 128000 - max_output_tokens: 2048 - require_max_tokens: true + input_price: 5.32 + output_price: 16 + supports_function_calling: true - name: meta.llama3-1-70b-instruct-v1:0 max_input_tokens: 128000 - max_output_tokens: 2048 - require_max_tokens: true input_price: 2.65 output_price: 3.5 + supports_function_calling: true - name: meta.llama3-1-8b-instruct-v1:0 max_input_tokens: 128000 - max_output_tokens: 2048 - require_max_tokens: true input_price: 0.3 output_price: 0.6 + supports_function_calling: true - name: meta.llama3-70b-instruct-v1:0 max_input_tokens: 8192 - max_output_tokens: 2048 - require_max_tokens: true input_price: 2.65 output_price: 3.5 - name: meta.llama3-8b-instruct-v1:0 max_input_tokens: 8192 - max_output_tokens: 2048 - require_max_tokens: true input_price: 0.3 output_price: 0.6 - - name: mistral.mistral-large-2402-v1:0 - max_input_tokens: 32000 - max_output_tokens: 8192 - require_max_tokens: true - input_price: 8 - output_price: 2.4 - - name: mistral.mixtral-8x7b-instruct-v0:1 - max_input_tokens: 32000 - max_output_tokens: 8192 - require_max_tokens: true - input_price: 0.45 - output_price: 0.7 + - name: mistral.mistral-large-2407-v1:0 + max_input_tokens: 128000 + input_price: 3 + output_price: 9 + supports_function_calling: true + - name: cohere.command-r-plus-v1:0 + max_input_tokens: 128000 + input_price: 3 + output_price: 15 + supports_function_calling: true + - name: cohere.command-r-v1:0 + max_input_tokens: 128000 + input_price: 0.5 + output_price: 1.5 + supports_function_calling: true + - name: cohere.embed-english-v3 + type: embedding + max_input_tokens: 512 + input_price: 0.1 + output_vector_size: 1024 + default_chunk_size: 1000 + max_batch_size: 96 + - name: cohere.embed-multilingual-v3 + type: embedding + max_input_tokens: 512 + input_price: 0.1 + output_vector_size: 1024 + default_chunk_size: 1000 + max_batch_size: 96 - platform: cloudflare # docs: |