fix(local-models): use pm pins and resumable component downloads
Use PM for engine binaries, dependent runtime archives, and model files. Keep one operation-owned pause event through installation and download. Report whole-plan bytes and retain paused jobs across desktop remounts. Preserve the completed lm-pm worktree as its own integration parent. The owning session verified focused Python and desktop tests, actual Windows ARM64 CUDA downloads, and rendered pause/resume controls. Combined verification with the safety repairs follows in the merge.
This commit is contained in:
@@ -1102,6 +1102,16 @@ class LlamaCppVulkan(LlamaCpp):
|
||||
}
|
||||
|
||||
|
||||
@register
|
||||
class LlamaCppHip(LlamaCpp):
|
||||
name = "llamacpp-hip"
|
||||
backend = "hip"
|
||||
assets = {
|
||||
"win32-x64": "win-rocm-7.14-x64",
|
||||
"linux-x64": "ubuntu-rocm-7.14-x64",
|
||||
}
|
||||
|
||||
|
||||
@register
|
||||
class LlamaCppMetal(LlamaCpp):
|
||||
"""macOS archives are unified builds with Metal compiled in."""
|
||||
|
||||
Reference in New Issue
Block a user