feat: add adaptive GPU scheduling

modelhub_submmit_api/README.md

@@ -8,6 +8,7 @@ It currently supports:
 - continuous queue refill via `run_poll.sh`
 - multiple ModelHub tokens read from `KEY.md` and `KEYS.md`
 - automatic task/framework/template selection across the supported GPU catalog
+- adaptive long-term/exploration/recent GPU scheduling with a persistent local snapshot
 
 ## Layout
 
@@ -72,13 +73,21 @@ bash run_poll.sh --dry-run
 ## Behavior
 
 - The runner auto-discovers all safe GPU/template combinations from the public submit catalog.
+- Automatic GPU selection uses smooth 50/30/20 scheduling: long-term Wilson-ranked top 3 GPUs,
+  all supported GPUs, and the best GPU from the latest 1,000 terminal tasks.
+- A strategy generation lasts exactly 200 platform-accepted submissions. Rejected API calls and
+  duplicates do not advance it. The next cycle refreshes platform history before submitting again.
+- Strategy state is stored in `.modelhub_state/gpu_strategy.json`; a generation never recalculates
+  during candidate submission.
 - Each model can be submitted at most once per GPU.
 - Multiple ModelHub tokens are pooled and used to route submissions to the account with available async capacity.
 - Concurrent submissions reserve account slots locally, and an account-capacity race automatically falls through to another account.
 - Concurrent local processes claim model/GPU pairs in `.modelhub_state/submission_claims.jsonl`; shared ledger, history, and outcome files use process locks and atomic replacement.
-- Isolated agent containers diversify candidate order by instance identity to reduce cross-container duplicate submissions.
-- History stats are online-only. The local ledger is used for local accounting, but platform history is only used after the local ledger reaches the configured threshold.
-- The default history threshold is `500` records.
+- ModelScope list pages are paced and cached for 15 minutes. HTTP 429 responses use exponential
+  backoff and `Retry-After`; pages already downloaded remain usable and the failed page is retried
+  on the next cycle.
+- Every third poll cycle, a full account gets one controlled capacity probe. A successful probe
+  raises that account's persisted known limit; a capacity rejection enters cooldown.
 
 ## Important Flags
 
@@ -95,6 +104,8 @@ Common flags:
 - `--skip-outcome-sync`: skip outcome sync before scanning
 - `--skip-history-archive`: skip history archive download for this run
 - `--dry-run`: plan only, do not submit
+- `--gpu-strategy-refresh-submissions`: accepted tasks per strategy generation (default `200`)
+- `--disable-gpu-strategy`: restore legacy ordering; explicit `--gpu/--gpus` also bypasses adaptive selection
 - `run_daily.sh` injects `--daily-target 3` when no daily-target flag is provided. Set `SUBMIT_DAILY_TARGET` or pass `--daily-target` explicitly for a different target.
 
 `run_poll.sh` adds:
@@ -127,6 +138,11 @@ Each run typically includes:
 - `skipped.jsonl`
 - `failed.jsonl`
 
+Persistent local scheduler state is written under `.modelhub_state/`:
+
+- `gpu_strategy.json`: GPU ranks, generation progress, and 50/30/20 accepted counters
+- `account_capacity.json`: learned per-account active-task limits
+
 ## Verification
 
 Run the full test suite: