[alexcastgonz (pcbeingused333)]

Motivation and Context

Ollama reports why generation stopped in done_reason, but the Ollama chat connector never put it on the returned content, so finish_reason was always None. A reply cut off by the token limit (done_reason == "length") could not be told apart from a complete one. Separately, _get_metadata_from_chat_response dropped the usage whenever prompt_eval_count was 0 (a fully cached prompt), because it tested truthiness instead of is not None.

Description

• OllamaChatCompletion._get_finish_reason maps done_reason: length -> FinishReason.LENGTH; stop -> FinishReason.STOP, or FinishReason.TOOL_CALLS when the message has tool calls (Ollama reports stop for those, other connectors report tool_calls); anything else (including a streaming chunk that is not the last one) -> None.
• It is applied on all four paths: ChatResponse and mapping responses, streaming and not.
• The ChatResponse usage check now uses is not None, so prompt_eval_count == 0 keeps the usage.
• Tests (tests/unit/connectors/ai/ollama/services/test_ollama_chat_completion.py): finish reason for stop / length / stop with tool calls / unknown / missing on real ollama ChatResponse objects and on mappings; a streamed response where only the last chunk carries length; usage with prompt_eval_count == 0. These fail on main and pass with the change; the full Ollama unit test directory passes and ruff check / ruff format --check are clean. Not run against a live Ollama server.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

This change was made with Claude Code (Claude Sonnet 5.5).

[Copilot]

🟡 Changes recommended

Both streaming converters incorrectly report STOP instead of TOOL_CALLS when tool calls precede an empty terminal chunk, requiring a fix and regression coverage for both response types.

1 open issue

• Preserve TOOL_CALLS when terminal chunk follows tool calls · New

What changed in this PR

This PR updates the Python Ollama chat connector to expose generation finish reasons and preserve usage data when prompt token counts are zero.

Changes:

• Maps Ollama’s done_reason to finish reasons for typed and mapping responses, including streaming.
• Preserves token usage when counts are zero rather than missing.
• Adds tests for finish reasons, terminal streaming chunks, and zero-prompt-token usage.

File	Description
python/​tests/​unit/​connectors/​ai/​ollama/​services/​test_ollama_chat_completion.py	Adds finish-reason and zero-token usage tests; updates response doubles.
python/​semantic_kernel/​connectors/​ai/​ollama/​services/​ollama_chat_completion.py	Adds finish-reason mapping across response paths and retains zero-token usage.

_{Review effort: Balanced}