Merge pull request #265805 from aahill/aoai-param-revert

removing chunk size

Author: prmerger-automator[bot]

articles/ai-services/openai/concepts/use-your-data.md

@@ -232,15 +232,6 @@ When you want to reuse the same URL/web address, you can select [Azure AI Search
 ---
 
 
-## Ingestion parameters
-
-You can use the following parameter to change how your data is ingested in Azure OpenAI Studio, Azure AI Studio, and the ingestion API. Changing the parameter requires re-ingesting your data into Azure Search.
-
-|Parameter name  | Description  |
-|---------|---------|
-| **Chunk size** | Azure OpenAI on your data processes your documents by splitting them into chunks before indexing them in Azure Search. The chunk size is the maximum number of tokens for any chunk in the search index. The default chunk size is 1024 tokens. However, given the uniqueness of your data, you might find a different chunk size (such as 256, 512, or 1,536 tokens for example) more effective. Adjusting the chunk size can enhance the performance of the chat bot. While finding the optimal chunk size requires some trial and error, start by considering the nature of your dataset. A smaller chunk size is generally better for datasets with direct facts and less context, while a larger chunk size might be beneficial for more contextual information, though it can affect retrieval performance. This is the `chunkSize` parameter in the API.|
-
-
 ## Runtime parameters
 
 You can modify the following additional settings in the **Data parameters** section in Azure OpenAI Studio and [the API](../reference.md#completions-extensions). You do not need to re-ingest your data when you update these parameters. 
@@ -249,7 +240,7 @@ You can modify the following additional settings in the **Data parameters** sect
 |Parameter name  | Description  |
 |---------|---------|
 | **Limit responses to your data** | This flag configures the chatbot's approach to handling queries unrelated to the data source or when search documents are insufficient for a complete answer. When this setting is disabled, the model supplements its responses with its own knowledge in addition to your documents. When this setting is enabled, the model attempts to only rely on your documents for responses. This is the `inScope` parameter in the API. |
-|**Top K Documents**     |  This parameter is an integer that can be set to 3, 5, 10, or 20, and controls the number of document chunks provided to the large language model for formulating the final response. By default, this is set to 5. The search process can be noisy and sometimes, due to chunking, relevant information may be spread across multiple chunks in the search index. Selecting a top-K number, like 5, ensures that the model can extract relevant information, despite the inherent limitations of search and chunking. However, increasing the number too high can potentially distract the model. Additionally, the maximum number of documents that can be effectively used depends on the version of the model, as each has a different context size and capacity for handling documents. If you find that responses are missing important context, try increasing this parameter. Conversely, if you think the model is providing irrelevant information alongside useful data, consider decreasing it. When experimenting with the [chunk size](#ingestion-parameters), we recommend adjusting the top-K parameter to achieve the best performance. Usually, it is beneficial to change the top-K value in the opposite direction of your chunk size adjustment. For example, if you decrease the chunk size from the default of 1024, you might want to increase the top-K value to 10 or 20. This ensures a similar amount of information is provided to the model, as reducing the chunk size decreases the amount of information in the 5 documents given to the model. This is the `topNDocuments` parameter in the API. |
+|**Top K Documents**     |  This parameter is an integer that can be set to 3, 5, 10, or 20, and controls the number of document chunks provided to the large language model for formulating the final response. By default, this is set to 5. The search process can be noisy and sometimes, due to chunking, relevant information may be spread across multiple chunks in the search index. Selecting a top-K number, like 5, ensures that the model can extract relevant information, despite the inherent limitations of search and chunking. However, increasing the number too high can potentially distract the model. Additionally, the maximum number of documents that can be effectively used depends on the version of the model, as each has a different context size and capacity for handling documents. If you find that responses are missing important context, try increasing this parameter. Conversely, if you think the model is providing irrelevant information alongside useful data, consider decreasing it. This is the `topNDocuments` parameter in the API. |
 | **Strictness**     | Determines the system's aggressiveness in filtering search documents based on their similarity scores. The system queries Azure Search or other document stores, then decides which documents to provide to large language models like ChatGPT. Filtering out irrelevant documents can significantly enhance the performance of the end-to-end chatbot. Some documents are excluded from the top-K results if they have low similarity scores before forwarding them to the model. This is controlled by an integer value ranging from 1 to 5. Setting this value to 1 means that the system will minimally filter documents based on search similarity to the user query. Conversely, a setting of 5 indicates that the system will aggressively filter out documents, applying a very high similarity threshold. If you find that the chatbot omits relevant information, lower the filter's strictness (set the value closer to 1) to include more documents. Conversely, if irrelevant documents distract the responses, increase the threshold (set the value closer to 5). This is the `strictness` parameter in the API. |
 
 

articles/ai-services/openai/whats-new.md

@@ -50,10 +50,6 @@ Azure OpenAI Service now supports text to speech APIs with OpenAI's voices. Get
 - [Fine-tuning & function calling](./how-to/fine-tuning-functions.md)
 - [`gpt-35-turbo 1106` support](./concepts/models.md#fine-tuning-models)
 
-### Chunk size parameter for Azure OpenAI on your data
-
-- You can now set the [chunk size](./concepts/use-your-data.md#ingestion-parameters) parameter when your data is ingested. Adjusting the chunk size can enhance the model's responses by setting the maximum number of tokens for any given chunk of your data in the search index.
-
 ### New regional support for Azure OpenAI on your data
 
 You can now use Azure OpenAI on your data in the following Azure region: