Fix GPT-OSS Harmony format end token handling crash

[trvon]

Summary

Fix server crash when using GPT-OSS models with tools that was caused by unconsumed <|end|> tokens in the chat parser.

Changes

Module: chat-parser

• Problem: GPT-OSS server crashed with "Unexpected content at end of input" when tools were used
• Root Cause: common_chat_parse_gpt_oss() only consumed content when parse_tool_calls=false, leaving <|end|> token unconsumed when parse_tool_calls=true (tools enabled)
• Solution: Modified parser to consume all remaining content in both parse_tool_calls cases

Manual Validation

Before Fix:

# Server crash with GPT-OSS + tools
curl -X POST http://127.0.0.1:8080/v1/chat/completions \
  -d '{"messages": [{"role": "user", "content": "What is weather?"}], "model": "gpt-oss-20b", "tools": [...]}'
# Result: Server crash - "Unexpected content at end of input"

After Fix:

# Same request now succeeds
# Result: {"choices":[{"message":{"role":"assistant","reasoning_content":"...","content":""}...}]}

[Copilot]

Pull Request Overview

Fix server crash when using GPT-OSS models with tools that was caused by unconsumed <|end|> tokens in the chat parser.

• Modified GPT-OSS chat parser to properly handle <|end|> tokens in both tool and non-tool scenarios
• Added reasoning format parameter passing from templates to server configuration
• Added comprehensive test coverage for GPT-OSS Harmony format parsing edge cases

Reviewed Changes

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

Show a summary per file

File	Description
common/chat.h	Added reasoning_format field to common_chat_params struct for template-specific configuration
common/chat.cpp	Fixed GPT-OSS parser to consume remaining content and handle `<
tools/server/utils.hpp	Added reasoning_format parameter passing from chat templates to server parameters
tools/server/server.cpp	Updated server to use template-provided reasoning format with proper fallback logic
tests/test-chat-parser.cpp	Added comprehensive test suite for GPT-OSS Harmony format edge cases and token handling

[Copilot]

This line is unreachable because it's inside an else block that follows a return statement that was removed. The logic flow suggests this should be part of the else clause for the if (!builder.syntax().parse_tool_calls) condition.

[trvon]

I made changes that address this in subsequent commits.

[trvon]

I see and I am merging the changes from #15181 against my own in another branch.