Qwen Streaming LLM
Consumes transcript or context text and emits raw semantic assistant response deltas.
| Property | Value |
|---|---|
| Node type | qwen.llm_stream |
| Layer / kind | algorithm / transform |
| Capability | language.generation.cancellable_streaming |
Ports
| Port | Direction | Schema |
|---|---|---|
text_in |
Input Text | Prompt or turn context, streaming |
signal_in |
Input Signal | Cancels the active SSE response and clears stale results |
text_out |
Output Text | Assistant response deltas, streaming |
event_out |
Output Event | Response completion state |
Configuration
model defaults to qwen-flash; system_prompt defines behavior; temperature defaults to
0.6. Presentation batching belongs to a downstream Speech Formatter. HTTP SSE runs on
a background worker; callbacks only start requests or drain a bounded queue, so
on_signal can close an in-flight response immediately. The Node requests Runtime-managed
internal wakeups through its Context, so it exposes no Clock Node or tick_in Port.
max_results_per_wakeup defaults to 32.
The Node does not know about Voice Room or any client protocol. A project Node maps text_out and
event_out when a remote UI needs them.
The protocol follows Alibaba Cloud's OpenAI-compatible Qwen streaming output.