coolguy/coorl-lost-cities
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Make interleaved traversal the default

Browse Source 
Co-Authored-By: Codex <codex@openai.com>
This commit is contained in:
coolguy
2026-05-07 23:21:15 +09:00
co-authored by Codex
parent b9bbb4fcf7
commit 09bbe7ccf7
2 changed files with 17 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
configs/deep_cfr/default.yaml
+5 -2
View File
@@ -42,11 +42,14 @@ traversal:
store_strategy_on_traverser_nodes: true store_strategy_on_traverser_nodes: true
store_strategy_on_opponent_nodes: false store_strategy_on_opponent_nodes: false
num_workers: 8 num_workers: 8
worker_chunk_size: 8 worker_chunk_size: 64
progress_every_traversals: 10 progress_every_traversals: 0
endpoint_depth_bucket_width: 100 endpoint_depth_bucket_width: 100
endpoint_depth_bucket_max: 1000 endpoint_depth_bucket_max: 1000
inference_backend: local inference_backend: local
scheduler: interleaved
interleave_width: 64
interleave_max_batch: 128
regret_matching: regret_matching:
all_negative_fallback: argmax_tiebreak all_negative_fallback: argmax_tiebreak
docs/performance.md
+12 -4
View File
@@ -532,8 +532,16 @@ game-state throughput and is slower end-to-end than 8 interleaved CPU workers.
Important caveat: multi-traversal interleaving uses per-context RNG streams, so Important caveat: multi-traversal interleaving uses per-context RNG streams, so
exact recursive-batch RNG ordering is intentionally not preserved. The Phase 2 exact recursive-batch RNG ordering is intentionally not preserved. The Phase 2
single-traversal parity test matches recursive stats and sample target single-traversal parity test matches recursive stats and sample target
checksums under identical RNG seed. Longer learning-curve A/B is still required checksums under identical RNG seed. This means the default now favors the
before considering a default switch. measured traversal-speed win over byte-identical sample ordering. If future
learning curves show unexplained drift, first compare against the recursive
fallback:
```bash
--set traversal.scheduler=recursive \
--set traversal.worker_chunk_size=8 \
--set traversal.progress_every_traversals=10
```
Follow-up: `average_strategy` support was added after the initial Phase 3 Follow-up: `average_strategy` support was added after the initial Phase 3
network-opponent A/B so the interleaved path can run the actual default opponent network-opponent A/B so the interleaved path can run the actual default opponent
@@ -546,8 +554,8 @@ settings produced warm-up-excluded means:
| interleaved, default `average_strategy` | 10.61 | 4.85 | 28.3 | 64 | | interleaved, default `average_strategy` | 10.61 | 4.85 | 28.3 | 64 |
Run: `runs/2026-05-07_230419_option-b-interleaved-average-strategy-10i`. Run: `runs/2026-05-07_230419_option-b-interleaved-average-strategy-10i`.
This prepares the long-run default-config A/B, but it does not replace it: This follow-up unblocked making interleaved traversal the default. The default
learning-curve stability still has to be measured before any default switch. switch was made with the caveat above rather than waiting for a long-run A/B.
## Batched Traversal Inference: Design Decision (2026-05-07) ## Batched Traversal Inference: Design Decision (2026-05-07)