CVE-2026-111160

Description

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint. Attackers can send a single request whose parameters cause pickle.loads to fail in InferSamplingParams, killing the inference thread while /health stays up and all later requests hang until restart.

References

• https://github.com/ModelTC/LightLLM
• https://github.com/ModelTC/LightLLM/issues/1629
• https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/api_lightllm.py#L37-L43
• https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/core/objs/sampling_params.py#L356-L357
• https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/infer_batch.py#L484-L485