Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75822.1565500799,
-      "required_pages_per_second_upper_estimate": 0.0065720723602620485
+      "required_token_positions_per_second": 75731.85763252276,
+      "required_pages_per_second_upper_estimate": 0.00656424547894871
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9477.769568759988,
-      "required_pages_per_second_upper_estimate": 0.0037808239862613642
+      "required_token_positions_per_second": 9466.482204065345,
+      "required_pages_per_second_upper_estimate": 0.00377632128772353
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18955.539137519976,
-      "required_pages_per_second_upper_estimate": 0.004385030373370489
+      "required_token_positions_per_second": 18932.96440813069,
+      "required_pages_per_second_upper_estimate": 0.004379808107027945
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28433.308706279964,
-      "required_pages_per_second_upper_estimate": 0.0017314393221707545
+      "required_token_positions_per_second": 28399.446612196036,
+      "required_pages_per_second_upper_estimate": 0.0017293772983017206
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28433.308706279964,
-      "required_pages_per_second_upper_estimate": 0.0021888748126454947
+      "required_token_positions_per_second": 28399.446612196036,
+      "required_pages_per_second_upper_estimate": 0.002186268014906638
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28433.308706279964,
-      "required_pages_per_second_upper_estimate": 0.3530465339692311
+      "required_token_positions_per_second": 28399.446612196036,
+      "required_pages_per_second_upper_estimate": 0.35262608008984736
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3513271322889864,
-    "producer_p95_seconds": 0.20198439799969492,
-    "gpu_update_seconds": 0.9723806780845552,
-    "producer_batches_per_second": 2.846350048414261,
-    "gpu_batches_per_second": 1.0284038160546862,
-    "producer_headroom": 2.767735790143066,
+    "producer_batch_seconds": 0.1143375932577726,
+    "producer_p95_seconds": 0.2546618079995824,
+    "gpu_update_seconds": 0.9735400966625409,
+    "producer_batches_per_second": 8.74602981843,
+    "gpu_batches_per_second": 1.0271790586008405,
+    "producer_headroom": 8.514610714847807,
     "sustainable": true,
-    "buffer_seconds": 60.28760204124242,
+    "buffer_seconds": 60.35948599307754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 206330.7413149635,
-    "upload_bytes_per_second": 6532567.137948461,
+    "generation_bytes_per_second": 219235.59043132063,
+    "upload_bytes_per_second": 6474041.960848427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.27377395235709,
+    "estimated_full_shards_per_hour": 47.04285416321482,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.027329577999808,
-      "bytes": 11020380,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2178119280006285,
       "bytes": 3672261,
@@ -562,6 +556,12 @@
       "bytes": 875148,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.2982671780009696,
+      "bytes": 8839899,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 10800, "loss": 0.7142677288502455, "lr": 0.0003, "grad_norm": 0.16578387173052617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1579822809, "tokens_per_second": 149322.5279770065, "data_wait_s": 6.496000423794612e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 10900, "loss": 0.6950379349291325, "lr": 0.0003, "grad_norm": 0.15290236586700576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1594471529, "tokens_per_second": 148803.07798327558, "data_wait_s": 1.2014999811071903e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11000, "loss": 0.722653703764081, "lr": 0.0003, "grad_norm": 0.14722297485559593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1609008434, "tokens_per_second": 139375.47298112747, "data_wait_s": 8.352999429916963e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 11100, "loss": 0.7419378254562616, "lr": 0.0003, "grad_norm": 0.15071296947925078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1623604035, "tokens_per_second": 148108.5150613146, "data_wait_s": 1.1514999641804025e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24d210f18482d14b69a81956ce55fef0d29788a2ae6cbe64471bd95f9c526cd1
-size 230802
+oid sha256:a85fadbd4a052813f58d2cef92a38590c46d3e26eb1f71b89afaf523a83f5981
+size 232885