card: present beam 1024 only; one line on the fast default

README.md

@@ -39,6 +39,9 @@ each):
 python solve.py --facelets <54 characters>
 ```
 
+Use `--beam 1024` for full scrambles, which needs a GPU. The default, 64, is
+fast enough for a laptop, but it only solves short scrambles.
+
 `solve.py` has no other dependencies. It downloads `config.json` and
 `model.safetensors` from this repo the first time.
 
@@ -77,19 +80,16 @@ full scrambles from 16% to 96%.
 Canonical RubiksBench scrambles (the same cubes every language model on the
 board is tested on), 25 cubes per scramble length:
 
-| Scramble length | beam 64 | beam 1024 |
-|---|---|---|
-| 3 | 100% | 100% |
-| 6 | 100% | 100% |
-| 10 | 100% | 100% |
-| 15 | 28% | 100% |
-| 20 | 24% | 96% |
+| Scramble length | Solved (beam 1024) |
+|---|---|
+| 3 | 100% |
+| 6 | 100% |
+| 10 | 100% |
+| 15 | 100% |
+| 20 | 96% |
 
 Past about 15 moves, every scramble is fully mixed, so 25, 50 or 100 moves are
-no harder than 20. Beam 1024 takes about 30 seconds per full scramble on one
-consumer graphics card. Beam 64 on a processor with no graphics card took about 5 seconds for a
-3-move scramble and about 50 seconds for a 6-move scramble; it is not practical
-for full scrambles.
+no harder than 20. A full scramble takes about 80 to 90 seconds on one RTX A6000.
 
 ## Model
 
@@ -103,7 +103,6 @@ for full scrambles.
 
 | What you did | What you get at 20 moves |
 |---|---|
-| Beam 64 instead of 1024 | about 24% |
 | Always take the best-scored move, no search | 34% even at 10 moves |
 | Judge the network by its average error | about 1.5 moves, which looks poor; the search only needs the correct ranking of the 18 neighbours |