Select your cookie preferences

We use essential cookies and similar tools that are necessary to provide our site and services. We use performance cookies to collect anonymous statistics, so we can understand how customers use our site and make improvements. Essential cookies cannot be deactivated, but you can choose “Customize” or “Decline” to decline performance cookies.

If you agree, AWS and approved third parties will also use cookies to provide useful site features, remember your preferences, and display relevant content, including relevant advertising. To accept or decline all non-essential cookies, choose “Accept” or “Decline.” To make more detailed choices, choose “Customize.”
Sign in
Sign in
or
Create a new account
About
Categories
Delivery Methods
Solutions
Resources
 Your Saved List Become a Channel Partner Sell in AWS Marketplace Amazon Web Services Home Help

Amazon Sagemaker

Amazon SageMaker is a fully-managed platform that enables developers and data scientists to quickly and easily build, train, and deploy machine learning models at any scale. With Amazon SageMaker, all the barriers and complexity that typically slow down developers who want to use machine learning are removed. The service includes models that can be used together or independently to build, train, and deploy your machine learning models.

product logo

Cohere Command R Fine-tuning Free trial

By:
Cohere
Latest Version:
v1.0.0
A fine-tunable version of Command R. Command R is a generative model optimized for long-context tasks and large scale production workloads.
Continue to Subscribe

    Product Overview

    Command R is a highly performant generative large language model, optimized for a variety of use cases including reasoning, summarization, and question answering. Command R has the capability for multilingual generation evaluated in 10 languages and highly performant RAG capabilities. It is built for enterprises that plan to leverage their internal data and documents for a tailored and accurate language model. This instance is fine-tunable, allowing for customization on advanced use cases by leveraging your data. To access Cohere's Command R Finetuning model, please refer to the Sagemaker listing as Jumpstart is currently not supporting finetuning capabilities. Batch transform is not supported with this model.

    Key Data

    Version
    v1.0.0
    Usage Instructions
    Release Notes
    By
    Cohere
    Categories
    Natural Language Processing
    Generation-Text
    Summarization-Text
    Type
    Algorithm
    Fulfillment Methods
    Amazon SageMaker

    Highlights

    • Command R with fine-tuning allows you to customize your models to be performant for your business, domain, and industry. Alongside the fine-tuned model, users additionally benefit from Command R’s proficiency in the most commonly used business languages (10 languages) and retrieval-augmented generation (RAG) with citations for accurate and verified information.

    • Command R with fine-tuning achieves high levels of performance with less resource usage on targeted use cases. Enterprises will see lower operational costs, improved latency and increased throughput without extensive computational demands.

    • It excels at tasks such as: document summarization, content Q&A, long-form generation, and content generation amongst others. With fine-tuning it can power industry and business specific knowledge assistants, chatbots, customer support agents and more.

    Not quite sure what you’re looking for? AWS Marketplace can help you find the right solution for your use case. Contact us

    Pricing Information

    Use this tool to estimate the software and infrastructure costs based your configuration choices. Your usage and costs might be different from this estimate. They will be reflected on your monthly AWS billing reports.

    Contact us to request contract pricing for this product.


    Estimating your costs

    Choose your region and launch option to see the pricing details. Then, modify the estimated price by choosing different instance types.

    Version
    Region

    Software Pricing

    Algorithm Training$32.89/hr

    running on ml.p4de.24xlarge

    Model Realtime Inference$32.89/hr

    running on ml.p4de.24xlarge

    Model Batch Transform$32.89/hr

    running on ml.g4dn.12xlarge

    Infrastructure Pricing

    With Amazon SageMaker, you pay only for what you use. Training and inference is billed by the second, with no minimum fees and no upfront commitments. Pricing within Amazon SageMaker is broken down by on-demand ML instances, ML storage, and fees for data processing in notebooks and inference instances.
    Learn more about SageMaker pricing

    SageMaker Algorithm Training$47.1106/host/hr

    running on ml.p4de.24xlarge

    SageMaker Realtime Inference$47.111/host/hr

    running on ml.p4de.24xlarge

    SageMaker Batch Transform$4.89/host/hr

    running on ml.g4dn.12xlarge

    About Free trial

    Try this product for 7 days. There will be no software charges, but AWS infrastructure charges still apply. Free Trials will automatically convert to a paid subscription upon expiration.

    Algorithm Training

    For algorithm training in Amazon SageMaker, the software is priced based on hourly pricing that can vary by instance type. Additional infrastructure cost, taxes or fees may apply.
    InstanceType
    Algorithm/hr
    ml.p4de.24xlarge
    Vendor Recommended
    $32.89

    Usage Information

    Training

    To train a custom model, please see the example below for parameters to pass to co.finetuning.create_finetuned_model(), or visit our [API guide] . You can read about the Hyperparameters to tune here

    Channel specification

    Fields marked with * are required

    training

    *
    Input modes: File
    Content types: -
    Compression types: None

    evaluation

    Input modes: File
    Content types: -
    Compression types: None

    Hyperparameters

    Fields marked with * are required

    name

    *
    Name of the custom model
    Type: FreeText
    Tunable: No

    train_epochs

    Maximum number of training epochs to run for
    Type: Integer
    Tunable: No

    learning_rate

    The learning rate
    Type: Continuous
    Tunable: No

    train_batch_size

    Batch size to use used during training. Should be a multiple of 8
    Type: Integer
    Tunable: No

    early_stopping_enabled

    When enabled, stop traininig if the loss metric does not improve beyond 'early_stopping_threshold' for `early_stopping_patience` times of evaluations
    Type: Categorical
    Tunable: No

    early_stopping_patience

    Stop traininig if the loss metric does not improve beyond 'early_stopping_threshold' for this many times of evaluations
    Type: Integer
    Tunable: No

    early_stopping_threshold

    How much the loss must improve to prevent early stopping
    Type: Continuous
    Tunable: No

    Model input and output details

    Input

    Summary

    The model accepts JSON requests with parameters that can be used to control the generated text. See examples and fields descriptions below.

    Input MIME type
    application/json
    Sample input data
    Expand all input descriptions
    Field name
    message
    Description
    Text input for the model to respond to.
    Required
    Yes
    Data type
    FreeText
    Field name
    chat_history
    Description
    chat_history – (array of messages) A list of previous messages between the user and the model, meant to give the model conversational context for responding to the user's message. Required fields: role – (enum string) Takes “USER” or “CHATBOT”. message – (string) Text contents of the message.
    Required
    No
    Data type
    FreeText
    Default value
    []
    Field name
    documents
    Description
    A list of texts that the model can cite to generate a more accurate reply. Each document is a string-string dictionary. The resulting generation will include citations that reference some of these documents. It is recommended to keep the total word count of the strings in the dictionary to under 300 words. An `_excludes` field (array of strings) can be optionally supplied to omit some key-value pairs from being shown to the model.
    Required
    No
    Data type
    FreeText
    Default value
    []
    Field name
    search_queries_only
    Description
    When `true`, the response will only contain a list of generated search queries, but no search will take place, and no reply from the model to the user's `message` will be generated.
    Required
    No
    Data type
    Categorical
    Default value
    FALSE
    Allowed values
    TRUE, FALSE
    Field name
    preamble
    Description
    Overrides the default preamble for search query generation. Has no effect on tool use generations.
    Required
    No
    Data type
    FreeText
    Default value
    []
    Field name
    stream
    Description
    When `true`, the response will be a JSON stream of events. The final event will contain the complete response, and will have an `event_type` of `"stream-end"`. Streaming is beneficial for user interfaces that render the contents of the response piece by piece, as it gets generated.
    Required
    No
    Data type
    Categorical
    Default value
    FALSE
    Allowed values
    TRUE, FALSE
    Field name
    max_tokens
    Description
    The maximum number of tokens the model will generate as part of the response. Note: Setting a low value may result in incomplete generations.
    Required
    No
    Data type
    Integer
    Range
    Min: 0
    Default value
    []
    Field name
    temperature
    Description
    Use a lower value to decrease randomness in the response. Randomness can be further maximized by increasing the value of the `p` parameter.
    Required
    No
    Data type
    Continuous (Float)
    Range
    Min: 0
    Max: 2
    Default value
    0.3
    Field name
    Top P (p)
    Description
    Use a lower value to ignore less probable options. Set to 0 or 1.0 to disable. If both p and k are enabled, p acts after k.
    Required
    No
    Data type
    Continuous (Float)
    Range
    Min: 0.01
    Max: 0.99
    Default value
    0.75
    Field name
    Top K (k)
    Description
    Specify the number of token choices the model uses to generate the next token. If both p and k are enabled, p acts after k.
    Required
    No
    Data type
    Continuous (Float)
    Range
    Min: 0
    Max: 500
    Default value
    0
    Field name
    prompt_truncation
    Description
    Dictates how the prompt will be constructed. With `prompt_truncation` set to "AUTO_PRESERVE_ORDER", some elements from `chat_history` and `documents` will be dropped to construct a prompt that fits within the model's context length limit. During this process the order of the documents and chat history will be preserved. With `prompt_truncation` set to "OFF", no elements will be dropped.
    Required
    No
    Data type
    Categorical
    Default value
    OFF
    Allowed values
    AUTO_PRESERVE_ORDER, AUTO, OFF
    Field name
    frequency_penalty
    Description
    Used to reduce repetitiveness of generated tokens. The higher the value, the stronger a penalty is applied to previously present tokens, proportional to how many times they have already appeared in the prompt or prior generation.
    Required
    No
    Data type
    Continuous (Float)
    Range
    Min: 0.0
    Max: 1.0
    Default value
    0.0
    Field name
    presence_penalty
    Description
    Used to reduce repetitiveness of generated tokens. Similar to `frequency_penalty`, except that this penalty is applied equally to all tokens that have already appeared, regardless of their exact frequencies.
    Required
    No
    Data type
    Continuous (Float)
    Range
    Min: 0.0
    Max: 1.0
    Default value
    0.0
    Field name
    seed
    Description
    If specified, the backend will make a best effort to sample tokens deterministically, such that repeated requests with the same seed and parameters should return the same result. However, determinism cannot be totally guaranteed.
    Required
    No
    Data type
    Integer
    Range
    Default value
    []
    Field name
    return_prompt
    Description
    Returns the full prompt that was sent to the model when True.
    Required
    No
    Data type
    Categorical
    Default value
    FALSE
    Allowed values
    TRUE, FALSE

    Output

    Summary

    The output is a JSON object that has the generated text along with likelihoods of tokens, if requested. See example json.

    Output MIME type
    application/json
    Sample output data
    Expand all output descriptions
    Field name
    response_id
    Description
    Unique identifier for chat completion
    Always returned
    No
    Data type
    FreeText
    Default value
    []
    Field name
    generation_id
    Description
    Unique identifier for chat completion, used with Feedback endpoint on Cohere’s platform
    Always returned
    No
    Data type
    FreeText
    Default value
    []
    Field name
    text
    Description
    Model’s response to chat message input
    Always returned
    No
    Data type
    FreeText
    Default value
    []
    Field name
    finish_reason
    Description
    complete – The completion reached the end of generation token error_toxic – could not be completed due to our content filters. error_limit – could not be completed because the model’s context limit was reached. error – could not be completed due to an error. user_cancel - could not be completed because it was aborted by the user. max_tokens – could not be completed because the user specified a max_tokens limit in the request and this limit was reached. May not result in best performance.
    Always returned
    No
    Data type
    FreeText
    Default value
    []
    Field name
    token_count
    Description
    Count of tokens used
    Always returned
    No
    Data type
    Integer
    Range
    Default value
    []
    Field name
    meta
    Description
    API usage data, including current version and billable tokens.
    Always returned
    No
    Data type
    FreeText
    Default value
    []

    Sample notebook

    Additional Resources

    End User License Agreement

    By subscribing to this product you agree to terms and conditions outlined in the product End user License Agreement (EULA)

    Support Information

    Cohere Command R Fine-tuning

    AWS Infrastructure

    AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.

    Learn More

    Refund Policy

    No refunds. Please contact support+aws@cohere.com for further assistance.

    Customer Reviews

    There are currently no reviews for this product.
    View all