What happened
vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).
Affected versions
vllm: before 0.29.0 (semver) Fixed: See vendor advisory.
Why it matters
Review the vendor advisory and exposure of the affected product to determine operational impact.
Detection & mitigation
- Apply vendor-provided updates or mitigations.
- Review affected product exposure and access logs.
Public PoC references
No public PoC reference has passed the current publication threshold.