Problemsby IbisTrouble with vLLM decoding
MLOps3 posts from 3 people3d active+22 posts in the last 7 days, 0 the 7 days before (rising)
Posts per day
The posts behind it
3, newest first[Bug]: Hybrid GDN models (Qwen3.5) with n-gram speculative decoding: greedy outputs go wrong when drafts are partly accepted; the GDN state misses the last accepted token's update...later steps read slots. So the output reveals which writes the state contains. For every failing request, replaying the oracle shows the state was **missing the write of a token that had been accepted earlier**, most often the token accepted in the step just before the first...vllm-project/vllm · DiaaAj1 repliesOpen ↗[Bug]: Model Runner V2 hangs during speculative decoding warmup with pipeline parallelism### Your current environment The output of python collect_env.py ### 🐛 Describe the bug **Description** Code version: based on `main` commit [`[de7eeaa87bc19ecf088472d5f4471cdc9a5ab995](https://github.com/vllm-project/vllm/commit/de7eeaa87bc19ecf088472d5f4471cdc9a5ab995)`](https...vllm-project/vllm · chenjr971 repliesOpen ↗Support Qwen3.x MTP (Multi-Token Prediction) for speculative decoding## Summary Following analysis of 1Cat-vLLM's MTP-based speculative decoding for Qwen3.6-35B-A3B, this issue proposes adding MTP (Multi-Token Prediction) speculative decoding support f...[truncated]ggml-org/llama.cpp · hernandez423 reactionsOpen ↗
Companies and products named
About this problem
Evidence
3 posts from 3 people in 2 places, about 1 a week over 50 days. Mostly on vllm-project/vllm, ggml-org/llama.cpp. Tools named alongside: NVIDIA, llama.cpp, GitHub, Oracle.
Frustration Frustration 0 of 3· Seen on GitHub issues
How it was grouped
Posts about vLLM with no known issue, sharing the word "decoding". First post Aug 30, 2026, latest Oct 18, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
Other vLLM problems
vLLM upgrades break working setupsRunning vLLM locally hits memory and speed limitsvLLM GPU access and drivers are painfulRAG on vLLM returns irrelevant or missing contextTrouble with vLLM installationvLLM runs out of context on real workFine-tuning vLLM is costly and unpredictableTrouble with vLLM performancevLLM runs out of memoryvLLM lock-in makes switching costlyvLLM users keep filing bugs
History
Trouble with vLLM decoding (3 posts)