Name and Version

Current head in repo

d8794ee (HEAD -> master, origin/master, origin/HEAD) examples: refactor diffusion generation (#22590)

./build/bin/llama-server --version
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 32109 MiB):
Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes, VRAM: 32109 MiB
version: 9022 (d8794ee)
built with GNU 13.3.0 for Linux x86_64

Operating systems

Linux

GGML backends

CUDA

Hardware

GPU : GeForce RTX 5090
CPU : AMD EPYC 9J14

Models

Qwen3.5-35B-A3B-Q6_K.gguf (https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF)

Problem description & steps to reproduce

When Qwen models are used with GitHub CoPilot, sometimes copilot fails to process response.

Recent PRs fixed quite a few issues, this one still occures. Few sample requests are attached from network catpure.

Common pattern seems to be :

model generates a Qwen-style <tool_call>...</tool_call> block, but these are streamed inside delta.reasoning_content instead of converting it to delta.tool_calls.

• stream ends with data: [DONE]
• delta.content is empty
• delta.tool_calls is absent
• delta.reasoning_content contains a complete <tool_call>...</tool_call>
• final finish_reason is "stop"

Expected response :

• structured delta.tool_calls
• final finish_reason should be "tool_calls"

row_1142_read_file_requirements.request.sanitized.json
row_1230_run_terminal_backend.request.sanitized.json
row_1256_read_file_logger_after_compaction.request.sanitized.json

First Bad Commit

Does not seem to be a regression

Relevant log output

No error in logs , sample output that causes CoPilot to fail :

data: {"choices":[{"finish_reason":null,"index":0,"delta":{"reasoning_content":"<tool_call>"}}],...}
data: {"choices":[{"finish_reason":null,"index":0,"delta":{"reasoning_content":"<function=read_file>"}}],...}
data: {"choices":[{"finish_reason":null,"index":0,"delta":{"reasoning_content":"<parameter=filePath>"}}],...}
data: {"choices":[{"finish_reason":null,"index":0,"delta":{"reasoning_content":"/tmp/example-workspace/example-project/backend/requirements.txt"}}],...}
data: {"choices":[{"finish_reason":null,"index":0,"delta":{"reasoning_content":"</tool_call>"}}],...}
data: {"choices":[{"finish_reason":"stop","index":0,"delta":{}}],...}
data: [DONE]