Update README.md

README.md

@@ -28,7 +28,7 @@ device_map = {
     "model.transformer.ln_f": 0,
     "model.transformer.ff_out": 1,
 }
-# For 2 x 24 GB, this works for *only* 38 or 39. Any higher or lower and it'll either only work for <= 1 token of output.
+# For 2 x 24 GB, this works for *only* 38 or 39. Any higher or lower and it'll either only work for 1 token of output or fail completely.
 switch_point = 38 # layer index to switch to second GPU
 device_map |= {f"model.transformer.blocks.{i}": 0 for i in range(0, switch_point)}
 device_map |= {f"model.transformer.blocks.{i}": 1 for i in range(switch_point, 80)}