Trim obsolete Metal platform capability glue (#602)
This PR is: - To trust vLLM 0.27 platform defaults for device capability and quantization validation. - To remove Metal’s fake SM80 device capability. - To delete a test that only asserted the fake capability shape. vLLM now returns `None` for non-CUDA device capability, and an empty `supported_quantization` list already means there is no platform-side quantization filter. The Metal overrides were redundant and the fake CUDA capability was misleading. Signed-off-by: Yuan Lik Xun <lxyuan0420@gmail.com>
L
Lik Xun Yuan (Lx) committed
2cb7e4b1851f176cfc80ed4d770e20be74cabb80
Parent: ea0dbe0
Committed by GitHub <noreply@github.com>
on 8/13/2026, 2:31:57 AM