RL: pure-Python inferencia natrenovanej siete

py -m rl.export vyexportuje checkpoint do rl/weights/neural-bot.json
(bit-exact float32, 1.3 MB) a rl/pure_net.py ho hra bez torch/numpy
(stdlib forward pass, ~16 ms/tah). Testy parity: logity aj akcie sa
zhoduju s torch, identicke trajektorie celych kol. Natrenovany model:
6.7-6.9 b/kolo proti vsetkym baseline-om (heuristika prekonana).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
timandClaude Fable 5 committed 2026-07-07 18:49:55 +02:00
1 parent 8f2449a408
commit 9a750756c5
4 files changed
+339

No files matched your search

+1
View File
@@ -0,0 +1 @@
{"obs_dim": 233, "hidden": 384, "weights": {"trunk0_w": {"shape": [384, 233], "data": "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 truncated