feat: add adaptive GPU scheduling

README.md

@@ -31,6 +31,8 @@ Optional tuning:
 - `MODELHUB_AGENT_POST_CYCLE_COOLDOWN_SECONDS` default `2`
 - `MODELHUB_AGENT_MAX_SUBMITS_PER_RUN` default `0` (fill all currently available slots)
 - `MODELHUB_AGENT_ACTIVE_TASK_CAP` default `100` per account
+- `MODELHUB_CAPACITY_PROBE_INTERVAL_CYCLES` default `3`
+- `MODELHUB_CAPACITY_STATE_PATH` default `.modelhub_state/account_capacity.json`
 - `MODELHUB_AGENT_ACTIVE_COUNTS_TTL_SECONDS` default `15`
 - `MODELHUB_AGENT_RESERVATION_TTL_SECONDS` default `120`
 - `MODELHUB_AGENT_INSTANCE_ID` optional stable worker identity used to spread concurrent agents across accounts and candidates
@@ -39,6 +41,27 @@ Optional tuning:
 - `MODELHUB_AGENT_MIN_DOWNLOADS`
 - `MODELHUB_AGENT_GPUS`
 - `MODELHUB_AGENT_EXTRA_ARGS`
+- `MODELHUB_GPU_STRATEGY_STATE_PATH` default `.modelhub_state/gpu_strategy.json`
+- `MODELSCOPE_PAGE_INTERVAL_SECONDS` default `0.25`
+- `MODELSCOPE_PAGE_CACHE_TTL_SECONDS` default `900`
+
+## Adaptive GPU Strategy
+
+When no explicit GPU override is supplied, the worker uses a local 50/30/20
+strategy generation:
+
+- 50%: the three long-term GPUs with the best Wilson lower confidence score and at least 100 terminal samples
+- 30%: round-robin exploration across every currently supported GPU
+- 20%: the best GPU among the latest 1,000 terminal tasks
+
+Only platform-accepted tasks count. After exactly 200 accepted tasks, the next
+poll cycle reloads all account history, generates a new immutable strategy snapshot,
+and resets the generation counters to 100/60/40 targets. The active snapshot and
+progress are stored in `.modelhub_state/gpu_strategy.json`.
+
+ModelScope HTTP 429 responses use exponential backoff and `Retry-After`. Successful
+pages remain cached, so a later cycle retries the failed page instead of restarting
+the whole pagination scan.
 
 ## Concurrent Agents
 
@@ -56,7 +79,7 @@ authority for account capacity and model/GPU uniqueness.
 If the platform reports that a model/GPU is already being validated, the claim
 is retained and the runner immediately draws replacement candidates from the
 same scan instead of retrying the duplicate every cycle. Startup logs and the
-health response expose `agent_version`; version `2026.08.02.2` or newer includes
+health response expose `agent_version`; version `2026.08.02.3` or newer includes
 this behavior.
 
 ## Deploy