|
About the V1 Feedback category
|
|
0
|
68
|
March 20, 2025
|
|
Clarification: EP All-to-All Communication Across TP×DP — Diagram Validation
|
|
1
|
138
|
March 22, 2026
|
|
vLLM v1 forces me to pre-allocate a huge, non-reclaimable GPU KV cache for long contexts, and none of the current offload or quantization options solve the resulting VRAM bloat without crippling speed.
|
|
5
|
1205
|
September 8, 2025
|
|
GLM4.5 V memory leak on inference
|
|
9
|
983
|
August 28, 2025
|
|
自定义多模态模型注册集成到vllm的问题一个问题反馈
|
|
1
|
329
|
August 7, 2025
|
|
Cudagraph in V1
|
|
3
|
1433
|
July 21, 2025
|
|
vLLM V1 - Default max CUDA graph size
|
|
2
|
2072
|
June 30, 2025
|
|
Dose vllm V1 support asynchronous scheduling?
|
|
3
|
1538
|
June 27, 2025
|
|
V1 has lower end-to-end performance than V0(--num-scheduler-steps=8)
|
|
6
|
664
|
June 12, 2025
|
|
[Spec Decode] Why does the acceptance rate look close to the paper, but the throughput is still not high?
|
|
1
|
606
|
April 21, 2025
|
|
Why V1 not support Mamba Models?
|
|
1
|
232
|
April 21, 2025
|
|
The new V1 way to ~--cpu-offload-gb
|
|
5
|
4138
|
April 13, 2025
|
|
Does the vLLM v1 support Speculative Decoding now?
|
|
4
|
464
|
March 25, 2025
|