Refresh GLM Q2TP and Q4TP performance evidence
Browse files
README.md
CHANGED
|
@@ -107,7 +107,7 @@ Neither the 108.27 GiB nor the 155.68 GiB file needs to fit in VRAM. CMF keeps w
|
|
| 107 |
in host memory and detects the available adapter budget. When dynamic pooling
|
| 108 |
is enabled, routed experts use a bounded global GPU pool: resident experts run
|
| 109 |
on the GPU, while cache misses are completed exactly on CPU and accumulated
|
| 110 |
-
into the same MoE result.
|
| 111 |
adapter use the same file on CPU.
|
| 112 |
|
| 113 |
Useful controls:
|
|
|
|
| 107 |
in host memory and detects the available adapter budget. When dynamic pooling
|
| 108 |
is enabled, routed experts use a bounded global GPU pool: resident experts run
|
| 109 |
on the GPU, while cache misses are completed exactly on CPU and accumulated
|
| 110 |
+
into the same MoE result. Automatic mode and systems without a supported
|
| 111 |
adapter use the same file on CPU.
|
| 112 |
|
| 113 |
Useful controls:
|