Qwen Code Nightly v0.15.0 Fixes Streaming Parser
💡Fixes streaming parser conflicts in qwen-code—key for reliable AI coding streams
⚡ 30-Second TL;DR
What Changed
Nightly release v0.15.0-nightly.20260423.d40fe7cdb
Why It Matters
This minor fix enhances reliability for streaming tool calls in multi-stream setups, benefiting developers using qwen-code for real-time AI coding tasks.
What To Do Next
Pull qwen-code nightly v0.15.0 if using StreamingToolCallParser for multi-stream tool calls.
Key Points
- •Nightly release v0.15.0-nightly.20260423.d40fe7cdb
- •Fixes StreamingToolCallParser scoping to per-stream basis
- •PR #3525 by @zhangxy-zju
- •Changelog from v0.15.0
🧠 Deep Insight
AI-generated analysis for this event — not the original article.
🔑 Enhanced Key Takeaways
- •The fix addresses a critical state-management bug where tool call parsing state was leaked across concurrent streams, potentially causing malformed JSON or incorrect function invocation parameters in multi-user environments.
- •This update is part of a broader effort by the Qwen team to stabilize their streaming infrastructure for agentic workflows, which rely heavily on reliable, low-latency tool call parsing.
- •The transition to per-stream scoping indicates a shift toward more robust asynchronous handling within the Qwen-Code inference pipeline, likely to support higher concurrency in production deployments.
📊 Competitor Analysis▸ Show
| Feature | Qwen-Code (Streaming) | DeepSeek-Coder | CodeLlama |
|---|---|---|---|
| Streaming Tool Calling | Per-stream isolated parser | Varies by implementation | Limited native support |
| Architecture | Transformer-based | Mixture-of-Experts | Transformer-based |
| License | Apache 2.0 / Custom | Custom | Llama 3 Community |
🛠️ Technical Deep Dive
- The StreamingToolCallParser was previously initialized at the Converter level, creating a singleton-like state that persisted across multiple request contexts.
- The fix involves refactoring the parser to be instantiated within the stream's lifecycle scope, ensuring that the internal buffer and state machine are reset for every new request.
- This change prevents cross-contamination of token sequences when multiple tool calls are processed concurrently by the same worker process.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Qwen (GitHub Releases: qwen-code) ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.