Model Limitations in Generative AI
Review the following model requirements for the OCI Generative AI custom and base models to get the most out of your models.
For key features of the pretrained base models, see Pretrained Foundational Models in Generative AI.
Matching Base Models to Clusters
Expand the following sections to review the dedicated AI cluster unit size and units that match each foundational model.
Some OCI Generative AI foundational pretrained base models supported for the dedicated serving mode are now deprecated and will retire no sooner than 6 months after the release of the 1st replacement model. You can host a base model, or fine-tune a base model and host the fine-tuned model on a dedicated AI cluster (dedicated serving mode) until the base model is retired. For dedicated serving mode retirement dates, see Retiring the Models.
| Base Model | Fine-Tuning Cluster | Hosting Cluster | Pricing Page Information | Request Cluster Limit Increase |
|---|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
|
|
|
|
|
|
Not available for fine-tuning |
|
|
|
|
|
|
|
|
You must request a limit increase to use the following resources:
Meta Llama Family
-
To host a Meta Llama 3.2 11B Vision model, you must request
dedicated-unit-llama2-70-countto increase by 1. -
To host a Meta Llama 3.2 90B Vision model, you must request
dedicated-unit-llama2-70-countto increase by 2. -
To host a Meta Llama 3.1 (70B) model, you must request
dedicated-unit-llama2-70-countto increase by 2. -
To fine-tune a Meta Llama 3.1 (70B) model, you must request
dedicated-unit-llama2-70-countto increase by 4. -
To host a Meta Llama 3.1 (405B) model, you must request
dedicated-unit-llama2-70-countto increase by 8.
Cohere Command R Family
-
To host a Cohere Command R model, you must request
dedicated-unit-small-cohere-countto increase by 1. -
To fine-tune a Cohere Command R model, you must request
dedicated-unit-small-cohere-countto increase by 8. -
To host a Cohere Command R 08-2024 model, you must request
dedicated-unit-small-cohere-countto increase by 1. -
To fine-tune a Cohere Command R 08-2024 model, you must request
dedicated-unit-small-cohere-countto increase by 8. -
To host a Cohere Command R+ model, you must request
dedicated-unit-large-cohere-countto increase by 2. -
To host a Cohere Command R+ 08-2024 model, you must request
dedicated-unit-large-cohere-countto increase by 2.
References: Service Limits for Generative AI and Request Cluster Limit Increase
| Base Model | Fine-tuning Cluster | Hosting Cluster | Pricing Page Product Name | Request Cluster Limit Increase |
|---|---|---|---|---|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
|
Not available for fine-tuning |
|
|
|
- Not Available on-demand: All OCI Generative AI foundational pretrained models supported for the on-demand serving mode that use the text generation and summarization APIs (including the playground) are now retired. We recommend that you use the chat models instead.
- Can be hosted on clusters: If you host a summarization or a generation model such as
cohere.commandon a dedicated AI cluster, (dedicated serving mode), you can continue to use that model until it's retired. These models, when hosted on a dedicated AI cluster are only available in US Midwest (Chicago). See Retiring the Models for retirement dates and definitions.
| Base Model | Fine-tuning Cluster | Hosting Cluster | Pricing Page Product Name | Request Cluster Limit Increase |
|---|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
Not available for fine-tuning |
|
|
|
The
cohere.command model supported for the on-demand serving mode is now retired and this model is deprecated for the dedicated serving mode. If you're hosting cohere.command on a dedicated AI cluster, (dedicated serving mode) for summarization, you can continue to use this hosted model replica with the summarization API and in the playground until the cohere.command model retires for the dedicated serving mode. These models, when hosted on a dedicated AI cluster are only available in US Midwest (Chicago). See Retiring the Models for retirement dates and definitions. We recommend that you use the chat models instead which offer the same summarization capabilities, including control over summary length and style.| Base Model | Fine-tuning Cluster | Hosting Cluster | Pricing Page Product Name | Request Cluster Limit Increase |
|---|---|---|---|---|
|
Not available for fine-tuning |
|
|
|
- Units for Fine-Tuning Clusters
- Creating a fine-tuning dedicated AI cluster automatically provisions a fixed number of units based on the base model: 8 units for
cohere.command-r-16kand 2 units for other models. You can't change this number, but you can use the same cluster to fine-tune several models. - Units for Hosting Clusters
-
- When creating a cluster, by default, one unit is created for the selected base model.
- You can increase throughput or requests per minute (RPM) by adding model replicas. For example, 2 replicas require 2 units. You can add model replicas when creating or editing a hosting cluster.
- Host up to 50 models on the same cluster, with the following restrictions:
- Host up to 50 of the same version of a fine-tuned or a pretrained model on the same cluster.
- Host different versions of the same base model, only if using
T-FEWfine-tuning method forcohere.commandandcohere.command-lightbase models.
Instead of committing to dedicated AI clusters, you can pay as you go for on-demand inferencing. With on-demand inferencing you reach the foundational models either through the Console, in the playground or through the API. For on-demand features, see Calculating Cost in Generative AI.
Adding Endpoints to Hosting Clusters
To host a model for inference on a hosting dedicated AI cluster, you must create an endpoint for that model. Then, you can add either add a custom model or a pretrained foundational model to that endpoint.
About Endpoint Aliases and Stack Serving
A hosting dedicated AI cluster can have up to 50 endpoints. Use these endpoints for the following use cases:
- Creating Endpoint Aliases
-
Create aliases with many endpoints. These 50 endpoints must either point to the same base model or the same version of a custom model. Creating many endpoints that point to the same model makes it easier to manage the endpoints, because you can use the endpoints for different users or different purposes.
- Stack Serving
-
Host several versions of a custom model on one cluster. This applies to
cohere.commandandcohere.command-lightmodels that are fine-tuned with theT-Fewtraining method. Hosting various versions of a fine-tuned model can help you to assess the custom models for different use cases.
To increase the call volume supported by a hosting cluster, you can increase its instance count.
Expand the following sections to review the requirements for hosting models on the same cluster.
Some OCI Generative AI foundational pretrained base models supported for the dedicated serving mode are now deprecated and will retire no sooner than 6 months after the release of the 1st replacement model. You can host a base model, or fine-tune a base model and host the fine-tuned model on a dedicated AI cluster (dedicated serving mode) until the base model is retired. For dedicated serving mode retirement dates, see Retiring the Models.
For hosting the pretrained base chat models, or fine-tuned chat models on a hosting dedicated AI cluster, use the following cluster unit size and endpoint rules that match each base model.
| Hosting Cluster Unit Size | Matching Rules |
|---|---|
Small Generic V2 for the base model, meta.llama-3.2-11b-vision-instruct |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the |
Large Generic for the base model, meta.llama-3.1-70b-instruct |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host several custom models on the same cluster:
|
Large Generic for the base model, meta.llama-3-70b-instruct |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host several custom models on the same cluster:
|
Large Generic V2 for the base model, meta.llama-3.2-90b-vision-instruct |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the |
Large Generic 4 for the base model, meta.llama-3.1-405b-instruct |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the |
Small Cohere V2 for the base model, cohere.command-r-16k |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host several custom models on the same cluster:
You can't host different versions of a custom model trained on the |
Small Cohere V2 for the base model, cohere.command-r-08-2024 |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host several custom models on the same cluster:
You can't host different versions of a custom model trained on the |
Large Cohere V2_2 for the base model, cohere.command-r-plus |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the |
Large Cohere V2_2 for the base model, cohere.command-r-plus-08-2024 |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the |
For hosting the embedding models on a hosting dedicated AI cluster, use the following cluster unit size and endpoint rules.
| Hosting Cluster Unit Size | Matching Rules |
|---|---|
Embed Cohere for the base models cohere.embed.english-light-v3.0, cohere.embed.english-v3.0, cohere.embed.multilingual-light-v3.0, and cohere.embed.multilingual-v3.0 |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models Fine-tuning not available for the Cohere Embed models. |
- Not Available on-demand: All OCI Generative AI foundational pretrained models supported for the on-demand serving mode that use the text generation and summarization APIs (including the playground) are now retired. We recommend that you use the chat models instead.
- Can be hosted on clusters: If you host a summarization or a generation model such as
cohere.commandon a dedicated AI cluster, (dedicated serving mode), you can continue to use that model until it's retired. These models, when hosted on a dedicated AI cluster are only available in US Midwest (Chicago). See Retiring the Models for retirement dates and definitions.
To host the text generation models on a hosting dedicated AI cluster, use the following cluster unit size and endpoint rules that match your base model.
| Hosting Cluster Unit Size | Matching Rules |
|---|---|
Small Cohere for the base model, cohere.command-light |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host different custom models on the same cluster:
|
Large Cohere for the base model, cohere.command |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host different custom models on the same cluster:
|
Llama2 70 for the base model, meta.llama-2-70b-chat |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
|
The
cohere.command model supported for the on-demand serving mode is now retired and this model is deprecated for the dedicated serving mode. If you're hosting cohere.command on a dedicated AI cluster, (dedicated serving mode) for summarization, you can continue to use this hosted model replica with the summarization API and in the playground until the cohere.command model retires for the dedicated serving mode. These models, when hosted on a dedicated AI cluster are only available in US Midwest (Chicago). See Retiring the Models for retirement dates and definitions. We recommend that you use the chat models instead which offer the same summarization capabilities, including control over summary length and style.To host the pretrained cohere.command summarization model on a hosting dedicated AI cluster, use the following cluster unit size and endpoint rules.
| Hosting Cluster Unit Size | Matching Rules |
|---|---|
Large Cohere for the base model, cohere.command |
Hosting Base Models To host the same pretrained base model through several endpoints on the same cluster:
Hosting Custom Models To host different custom models on the same cluster:
|
Training Data
Datasets for training custom models have the following requirements:
- A maximum of one fine-tuning dataset is allowed per custom model. This dataset is randomly split to a 80:20 ratio for training and validating.
- Each file must have at least 32 prompt/completion pair examples.
- The file format is
JSONL. - Each line in the
JSONLfile has the following format:{"prompt": "<a prompt>", "completion": "<expected response given the prompt>"}\n - The file must be stored in an OCI Object Storage bucket.
Learn about Training Data Requirements in Generative AI.
Input Data for Text Embeddings
Input data for creating text embeddings has the following requirements:
- You can add sentences, phrases, or paragraphs for embeddings either one phrase at a time, or by uploading a file.
- Only files with a
.txtextension are allowed. - If you use an input file, each input sentence, phrase, or paragraph in the file must be separated with a newline character.
- A maximum of 96 inputs are allowed for each run.
- Each input must be less than 512 tokens. If an input is too long, select whether to cut off the start or the end of the text to fit within the token limit by setting the Truncate parameter to Start or End. If an input exceeds the 512 token limit and the Truncate parameter is set to None, you get an error message.
Learn about Creating text embeddings in OCI Generative AI.