This is the issue for HAMi RoapMap for v2.8, you can simply reply this PR to submit your ideas

Next Version: v2.8.0
Estimated release: Dec-2025/Jan-2026
Tasks

• Support enflame sGPU new plan @zhaikangqi331
• Website Optimization /assign @ouyangluwei163 @Nimbus318
• DRA design /assign @Shouren
• HAMi for KAI scheduler
• Ascend support for volcano @DSFans2014
• Support multiple cambricon types (370,590,etc..)
• Reduce the repo size @archlitchi
• Optimize performance under high concurrency @archlitchi @Shouren
  - Fix potential task inconsistency between Fit and AddResource
  - Do Not schedule pod before fully initialized
  - (Optional) AddUsage is heavy, we should use Pod Events(onAddPod) to track the cluster overview

Bugs:

• hami-device-plugin fails to parse NUMA affinity "0-1" from "nvidia-smi topo -m“ #1363
• hami cannot be used with the latest torch's cuda graph #1360
• Why isn't nodelock set to lock based on nodename #1342
• HAMi Scheduler Not Trying to Schedule Previously Pending Workload for 5mins #1368
• HAMi Scheduler Throttling With High Number of Pod Submission #1367
• vLLM 0.9.2 fails on HAMi vGPU when exposing only the “index-1” device: memory profiling assertion (Initial free == Current free) and HAMI-core ... host pid is error! #1381

For history releases, please refer to:
v2.6-v2.7 - #923