Has anyone got this same CUDA error: an illegal memory access was encountered while using this model in vllm:v0.25.0?
2
#20 opened 12 days ago
by 
lzachary
vLLM 0.23.1rc1.dev774+g2b802ad5b.d20260706. . load this model encounter OOM !!
#18 opened 20 days ago
by 
xuefeizzzzzz
SGLang support?
➕ 5
1
#17 opened 23 days ago
by 
danielus
[need help] vllm 0.24.0 on dgx spark warn Your GPU does not have native support for FP4
➕ 1
3
#16 opened 23 days ago
by 
imshenshen
doubled TPS NVFP4 vs FP8
🔥 4
3
#15 opened 24 days ago
by 
sSeeMyRolexXx
Fallback to marlin kernel give wired/incorrect and sometime gable result
#14 opened 24 days ago
by 
sapjunior
this model hallucinates easily
1
#13 opened 24 days ago
by 
batsclamp
vLLM + HGX B200: Your GPU does not have native support for FP4 computation but FP4 quantization is being used.
👍 3
2
#12 opened 25 days ago
by 
cgelias
Works via Ollama, but the ~40 tok/s MTP speedup requires raw llama-server (Ollama has no --spec-type flag) - plus vLLM lockup findings on DGX Spark
1
#11 opened 25 days ago
by 
darkmatter2222
为什么全是！！！！！！！！！
5
#10 opened 25 days ago
by 
jiecong
Why does this say 18B?
1
#9 opened 25 days ago
by darkmatter2222
NVFP4 vs FP8 on an RTX PRO 6000 Blackwell (vLLM 0.24): sadly, FP4 is still not faster than FP8 - Qwen3.6‑27B
👀👍 13
1
#8 opened 25 days ago
by 
janreges3
CUDA error: an illegal memory access was encountered using vllm(sm120)
1
#7 opened 25 days ago
by 
shakhizat
Does it support MTP?
➕👍 3
4
#6 opened 26 days ago
by 
sharon8811
will it work on rtx 5090 32gb vram ?
10
#5 opened 26 days ago
by 
arunsahu44
nvidia/Qwen3.6-27B-NVFP4 outputs repeated "!" / "d" tokens on Blackwell SM120
➕ 1
6
#4 opened 26 days ago
by 
richgua
Could we include BF16 as well as FP8 in the benchmarks please?
➕👀 5
#3 opened 26 days ago
by 
haydonryan
Qwen-AgentWorld-35B-A3B-NVFP4?
👍 2
#2 opened 27 days ago
by 
nagelanping
Approved
👍 1
1
#1 opened 27 days ago
by 
jasionkajakub