fix: bug file uploading error

[Priyanka2-Microsoft]

Purpose

• ...
  This pull request introduces improved compatibility with GPT-5 deployments and updates default model parameters across several components. The most significant changes are focused on handling GPT-5's unique requirements, such as default temperature settings and the lack of support for streaming or legacy sampling parameters.

GPT-5 Compatibility and Model Parameter Handling:

• Updated AzureOpenAITextGenerator.cs to detect GPT-5 deployments and bypass legacy sampling parameters (like Temperature, NucleusSamplingFactor, etc.), using API defaults instead. For GPT-5, the code now uses the non-streaming API, as streaming is not supported. Non-GPT-5 deployments continue to use the streaming API and legacy parameters. [1] [2] [3] [4]

• In ChatHost.cs, changed the prompt execution settings to avoid explicitly setting the Temperature parameter, preventing compatibility issues with GPT-5 which requires the default value (Temperature=1.0).

Default Parameter Updates:

• Updated the default Temperature parameter to 1 (from 0) in both ImageContextDecoder.cs and KeywordExtractingHandler.cs, aligning with GPT-5's requirements and improving consistency across the codebase. [1] [2]

Does this introduce a breaking change?

• Yes
• No

Golden Path Validation

• I have tested the primary workflows (the "golden path") to ensure they function correctly without errors.

Deployment Validation

• I have validated the deployment process successfully and all services are running as expected with this change.

What to Check

Verify that the following are valid

• ...

Other Information

[Copilot]

Pull request overview

This PR updates the solution’s OpenAI/Azure OpenAI invocation paths to better support GPT‑5 deployments by avoiding incompatible request parameters (notably non‑default temperatures) and using non‑streaming APIs where GPT‑5 streaming isn’t supported.

Changes:

• Adjust Azure OpenAI text/chat generation to detect GPT‑5 deployments and conditionally omit legacy sampling parameters and streaming calls.
• Stop explicitly setting Temperature in the backend chat execution settings to avoid GPT‑5 incompatibilities.
• Update Kernel Memory handler defaults from Temperature=0 to Temperature=1 in a couple of prompt execution settings.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
App/kernel-memory/service/Core/Handlers/KeywordExtractingHandler.cs	Changes the handler’s prompt execution temperature default from 0 to 1.
App/kernel-memory/service/Core/DataFormats/Image/ImageContextDecoder.cs	Changes the image decoding prompt execution temperature default from 0 to 1.
App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs	Adds GPT‑5 detection and switches between streaming/non‑streaming plus parameter omission for GPT‑5.
App/backend-api/Microsoft.GS.DPS/API/ChatHost/ChatHost.cs	Removes explicit temperature from prompt execution settings and adds GPT‑5 compatibility note.

Comments suppressed due to low confidence (1)

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:194

• Same as above: the comment claims GPT-5 uses max_completion_tokens, but this implementation does not set an alternative token-limit parameter for GPT-5 (it only omits MaxTokens/sampling params). The comment should reflect the actual behavior to avoid confusion when debugging token truncation/cost issues.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming, so we use non-streaming API for GPT-5.
            if (!isGpt5Deployment)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (2)

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:136

• These comments say GPT-5 uses max_completion_tokens, but this implementation doesn't actually set any token limit for GPT-5 (it just omits MaxTokens). Consider rewording to match the current behavior and avoid implying max_completion_tokens is being sent.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming.

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:193

• These comments say GPT-5 uses max_completion_tokens, but the current code does not send any GPT-5-specific token limit (it only omits MaxTokens). Please reword so the comment reflects actual behavior, or implement an explicit GPT-5 token-limit parameter once supported by the SDK.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming, so we use non-streaming API for GPT-5.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (4)

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:219

• In the chat-completions path, TokenSelectionBiases are still applied even when isGpt5Deployment is true (the block above this GPT-5 branching runs unconditionally). This is inconsistent with the text-completions path (where biases are skipped for GPT-5) and can reintroduce GPT-5 request validation failures if logit bias is not supported.

            if (options.StopSequences is { Count: > 0 })
            {
                foreach (var s in options.StopSequences) { openaiOptions.StopSequences.Add(s); }
            }

            if (options.TokenSelectionBiases is { Count: > 0 })
            {
                foreach (var (token, bias) in options.TokenSelectionBiases) { openaiOptions.TokenSelectionBiases.Add(token, (int)bias); }
            }

            openaiOptions.Messages.Add(new ChatRequestSystemMessage(prompt));

            // GPT-5 does not support streaming - use non-streaming API
            if (isGpt5Deployment)
            {
                Response<ChatCompletions>? response = await this._client.GetChatCompletionsAsync(openaiOptions, cancellationToken).ConfigureAwait(false);

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:136

• The GPT-5 comment says the request should use max_completion_tokens instead of max_tokens, but this code never sets any max-completion-tokens equivalent for GPT-5 (and also leaves MaxTokens unset). This is misleading for future maintainers reading the code.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming.

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:193

• Same as above: this GPT-5 comment claims max_completion_tokens behavior, but the implementation does not set a GPT-5-specific max-token property. Consider updating the comment to match the implementation (or implement max_completion_tokens if the SDK supports it).

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming, so we use non-streaming API for GPT-5.

App/backend-api/Microsoft.GS.DPS/Prompts/Chat_SystemPrompt.txt:16

• These instructions have grammatical issues that can reduce prompt clarity (e.g., "in your every sentences", "Make a detail answer"). Consider tightening the wording while keeping the intent the same.

You should include citations in your every sentences. The [Content] section already contains citations - do NOT add duplicate citations.
Make a detail answer as much as possible up to over 4000 characters.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (4)

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:198

• In the chat-completions path, TokenSelectionBiases are still applied for GPT-5 deployments even though other sampling parameters are gated behind !isGpt5Deployment. This is inconsistent with the completion path (where biases are gated) and can reintroduce GPT-5 request rejections if logit bias is treated as a legacy sampling parameter.

            if (!isGpt5Deployment)
            {
                openaiOptions.MaxTokens = options.MaxTokens;
                openaiOptions.Temperature = (float)options.Temperature;
                openaiOptions.NucleusSamplingFactor = (float)options.NucleusSampling;

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:136

• The comment says GPT-5 uses max_completion_tokens instead of max_tokens, but the code does not set any GPT-5-specific max token field (it just omits MaxTokens). This makes the comment misleading for future maintainers—either implement the alternative field (if supported) or update the comment to reflect the actual behavior.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming.

App/kernel-memory/extensions/AzureOpenAI/AzureOpenAITextGenerator.cs:193

• The comment mentions GPT-5 using max_completion_tokens instead of max_tokens, but the code path doesn’t actually set a GPT-5-specific max token parameter (it only omits MaxTokens). This mismatch can confuse future updates; either set the correct parameter (if available in the SDK) or update the comment to match the implementation.

            // GPT-5 deployments use max_completion_tokens instead of max_tokens
            // and require Temperature = 1.0 (default), rejecting other legacy sampling parameters.
            // GPT-5 also does NOT support streaming, so we use non-streaming API for GPT-5.

App/backend-api/Microsoft.GS.DPS/Prompts/Chat_SystemPrompt.txt:16

• This line still has grammatical issues: “Make a detail answer … up to over 4000 characters” is awkward/contradictory. Consider rephrasing to a single clear limit.

Make a detail answer as much as possible up to over 4000 characters.