feat: prioritize proven GPU framework combinations

README.md

@@ -48,26 +48,26 @@ Optional tuning:
 - `MODELHUB_MARKET_QUEUE_REFRESH_SECONDS` default `600`
 - `MODELHUB_MARKET_FRAMEWORK_REFRESH_SECONDS` default `21600`
 - `MODELHUB_MARKET_THROUGHPUT_WINDOW_HOURS` default `6`
-- `MODELHUB_MARKET_FRAMEWORK_MIN_SAMPLES` default `100`
+- `MODELHUB_MARKET_FRAMEWORK_MIN_SAMPLES` default `300`
 - `MODELSCOPE_PAGE_INTERVAL_SECONDS` default `0.25`
 - `MODELSCOPE_PAGE_CACHE_TTL_SECONDS` default `900`
 - `MODELHUB_AGENT_VERIFY_CACHE_TTL_SECONDS` default `900`
 
 ## Adaptive GPU Strategy
 
-When no explicit GPU override is supplied, the worker uses a queue-aware 50/30/20
-strategy generation:
+When no explicit GPU override is supplied, the worker uses a success-first 70/30
+strategy generation with no self-funded exploration:
 
-- 50%: the three long-term GPUs with the best Wilson lower confidence score and at least 100 terminal samples
-- 30%: exploration across every currently supported GPU
-- 20%: the top recent GPUs among the latest 1,000 terminal tasks
+- 70%: the three long-term GPUs with the best Wilson lower confidence score and at least 100 terminal samples
+- 30%: the top recent GPUs among the latest 1,000 terminal tasks
+- 0%: unvetted/all-GPU exploration; community-wide results provide the exploration signal
 
-The 50/30/20 category ratio remains exact across accepted tasks. Inside each
+The 70/30 category ratio remains exact across accepted tasks. Inside each
 category, weighted fair scheduling combines the category's historical rank with
 live public market data:
 
-- estimated backlog hours (`waiting / recent completions per hour`) instead of raw queue length
-- recent public success quality, scored with a Wilson lower confidence bound
+- recent public success quality, scored with a strongly weighted Wilson lower confidence bound
+- estimated backlog hours (`waiting / recent completions per hour`) as a bounded tie-breaker
 - machine availability, running workers, and advertised concurrency
 - a circuit breaker for unavailable or apparently stalled GPU pools
 
@@ -79,11 +79,12 @@ snapshot, uses a retry backoff, and never blocks normal submissions.
 For each compatible model/GPU pair, the worker also ranks the GPU's supported
 frameworks using ModelHub's public aggregate `modelCount` and `successCount` data,
 then blends in the worker's own GPU+framework outcomes with a capped weight.
-Only frameworks with at least 100 samples receive statistical priority; the
-legacy safe framework order remains the fallback. Framework statistics refresh
+Only frameworks with at least 300 public samples and a safe Wilson lower bound
+are eligible. Framework statistics refresh
 every 6 hours, so they do not add per-model API traffic. Newly published
 frameworks are discovered automatically, but receive no novelty bonus: they can
-win only when their confidence-adjusted success score is better. A new framework
+win only when their confidence-adjusted success score beats the best incumbent by
+at least 10%. A new framework
 is eligible only after the authenticated official build-config endpoint returns
 a complete config that passes local structure, placeholder, framework-name, and
 GPU-parallelism validation. Valid official configs are cached and refreshed with
@@ -91,9 +92,14 @@ the framework snapshot; local templates remain the fail-safe fallback.
 
 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
+and resets the generation counters to 140/60 targets. The active snapshot and
 progress are stored in `.modelhub_state/gpu_strategy.json`.
 
+Five consecutive local failures open a 12-hour GPU/framework circuit breaker.
+A sub-20% success rate over the latest 20 terminal tasks opens a 6-hour breaker.
+Candidate shortages expand the model search window; they never unlock an
+unvetted GPU or framework.
+
 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.
@@ -139,6 +145,10 @@ adds fail-closed model/GPU prechecks and persistent uniqueness exclusions.
 Version `2026.08.04.1` adds queue/throughput intelligence, GPU health circuit
 breaking, weighted-fair scheduling, live framework/config discovery, and
 confidence-ranked public-plus-local framework selection.
+Version `2026.08.05.1` removes self-funded GPU exploration, switches accepted
+traffic to 70/30 long-term/recent exploitation, raises the public framework gate
+to 300 samples, makes success dominate queue pressure, and adds recent local
+GPU/framework circuit breakers.
 
 ## Deploy