[Sayan Shaw (sayanshaw24)]

Why

Qwen templates define native tool protocols and <tool_response> history, but Foundry's canonical projection bypasses those contracts on continuation turns. In real multi-turn requests this can make the model repeat an already completed call, exhaust its output in hidden reasoning, or fail to select legitimate pending work.

This is stacked on #1231, which provides template-qualified protocol detection. #1264 remains a separate narrow Qwen2 required/named compatibility change.

What

• Project qualified Qwen tools, assistant calls, and tool results through the model's native template.
• Support native function/parameter XML and tagged-JSON tool-call dialects.
• Track completed name+arguments pairs and reject exact repeats before publication.
• Permit the same function with different arguments.
• Retry an intercepted exact repeat once with completed argument objects excluded, while retaining declared-tool lookup and full JSON Schema validation.
• Recover native post-tool turns that consume their budget in hidden reasoning without producing visible text.
• Keep native activation template-qualified so legacy/canonical models do not change behavior.
• Harden tagged-JSON parsing against malformed envelopes, duplicate/extra fields, and partial publication.

Validation

Deterministic combined stack validation:

• Qwen2 focused compatibility: 10/10 passed.
• Native Qwen regression selection: 114/114 passed.
• SDK integration: 158/158 passed.
• Cache suite: 17 passed, 2 optional BYOM skips.
• Unit suite: 2,630 passed, 3 skipped; one unrelated host-lifecycle timeout reproduced independently.

Real-model validation:

• Qwen 3.8 27B CUDA: 16/16 Ignite scenarios passed across streaming and non-streaming.
• Qwen 3.8 Flash Next FP8 Engine: 16/16 Ignite scenarios passed across streaming and non-streaming.
• Covered automatic greeting/no-call, required and named choice, single-result final answers, serial different-tool work, same-tool/different-arguments work, parallel results, exact-repeat prevention, and failed-tool no-retry.

The exact repeated call observed before this change is intercepted and never published; no repeated call occurred in either real-model matrix.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
foundry-local	Ready	Preview	Nov 8, 2026 3:10am UTC

[Copilot]

🟡 Changes recommended

Unresolved critical issues can cause freed-memory reads and route valid tool calls as ordinary text, bypassing parsing, validation, and repeat checks.

14 open issues

• Avoid reading a string view after modifying its backing string · New
• Preserve token provenance when replaying buffered continuations · New
• Fall back when native XML cannot represent the tool schema · New
• Limit completed-call interception to qualified native models · New
• Do not infer request completion from tool-name coverage · New
• Preserve parameterless schemas when adding retry exclusions · New
• Switch native XML retries to canonical JSON consistently · New
• Keep reasoning recovery within the original output budget · New
• Honor explicit response guidance during native activation · New
• Honor required and forced tool choices in native prompts · New
• Scope completed calls to the latest user request · New
• Preserve JSON escaping when restoring templated values · New
• Require decoder-compatible framing when qualifying XML templates · New
• Use the string-aware scanner to find the closing marker · New

What changed in this PR

This PR extends Foundry Local’s C++ chat pipeline to use template-qualified native Qwen tool protocols for multi-turn continuations, building on the protocol detection work in #1231.

Changes:

• Adds native XML and tagged-JSON prompt projection, parsing, and configurable tool boundaries.
• Tracks completed calls and introduces repeat-call and hidden-reasoning recovery.
• Expands capability and continuation tests, and includes the Qwen2 required/named fenced-output compatibility path.

File	Description
sdk_v2/​cpp/​test/​internal_api/​toolcalling/​tool_call_stream_accumulator_test.cc	Tests tagged-JSON parsing and custom boundaries.
sdk_v2/​cpp/​test/​internal_api/​genai_model_instance_test.cc	Tests template-qualified native capabilities.
sdk_v2/​cpp/​test/​internal_api/​chat/​chat_template_test.cc	Tests fenced required/named prompt instructions.
sdk_v2/​cpp/​test/​internal_api/​chat/​chat_session_test.cc	Tests repeat interception, retries, and capability precedence.
sdk_v2/​cpp/​src/​inferencing/​generative/​toolcalling/​tool_call_context.h	Adds native protocol and continuation state.
sdk_v2/​cpp/​src/​inferencing/​generative/​toolcalling/​qwen_xml_tool_call_decoder.h	Declares tagged-JSON parsing and configurable boundaries.
sdk_v2/​cpp/​src/​inferencing/​generative/​toolcalling/​qwen_xml_tool_call_decoder.cc	Implements tagged-JSON parsing and generalized XML boundaries.
sdk_v2/​cpp/​src/​inferencing/​generative/​genai_model_instance.h	Exposes inferred tool-protocol capabilities.
sdk_v2/​cpp/​src/​inferencing/​generative/​genai_model_instance.cc	Qualifies protocols through rendered-template probes.
sdk_v2/​cpp/​src/​inferencing/​generative/​chat/​prepared_chat_prompt.cc	Projects native tool definitions and conversation history.
sdk_v2/​cpp/​src/​inferencing/​generative/​chat/​chat_template.cc	Adds fenced tool-only instructions.
sdk_v2/​cpp/​src/​inferencing/​generative/​chat/​chat_session.h	Declares tool-support resolution helper.
sdk_v2/​cpp/​src/​inferencing/​generative/​chat/​chat_session.cc	Integrates native parsing, completion tracking, and recovery.

_{Review effort: Balanced}