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": 75701.90557463889,
-      "required_pages_per_second_upper_estimate": 0.0065616493104841325
+      "required_token_positions_per_second": 75222.50911223565,
+      "required_pages_per_second_upper_estimate": 0.006520096440142246
     },
     {
       "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": 9462.738196829861,
-      "required_pages_per_second_upper_estimate": 0.003774827747259399
+      "required_token_positions_per_second": 9402.813639029457,
+      "required_pages_per_second_upper_estimate": 0.0037509229452008363
     },
     {
       "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": 18925.476393659723,
-      "required_pages_per_second_upper_estimate": 0.0043780758866646405
+      "required_token_positions_per_second": 18805.627278058913,
+      "required_pages_per_second_upper_estimate": 0.0043503509030427875
     },
     {
       "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": 28388.214590489584,
-      "required_pages_per_second_upper_estimate": 0.0017286933271096613
+      "required_token_positions_per_second": 28208.44091708837,
+      "required_pages_per_second_upper_estimate": 0.0017177460535990974
     },
     {
       "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": 28388.214590489584,
-      "required_pages_per_second_upper_estimate": 0.0021854033427834495
+      "required_token_positions_per_second": 28208.44091708837,
+      "required_pages_per_second_upper_estimate": 0.002171563867759645
     },
     {
       "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": 28388.214590489584,
-      "required_pages_per_second_upper_estimate": 0.3524866159714117
+      "required_token_positions_per_second": 28208.44091708837,
+      "required_pages_per_second_upper_estimate": 0.3502544286115496
     }
   ],
   "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": 0.16150199280463085,
-    "producer_p95_seconds": 0.3086342990027333,
-    "gpu_update_seconds": 0.9739252855043035,
-    "producer_batches_per_second": 6.191874060710205,
-    "gpu_batches_per_second": 1.02677280781574,
-    "producer_headroom": 6.030422712383878,
+    "producer_batch_seconds": 0.13889357808602654,
+    "producer_p95_seconds": 0.3480862360011088,
+    "gpu_update_seconds": 0.9801321555226804,
+    "producer_batches_per_second": 7.199756920227297,
+    "gpu_batches_per_second": 1.0202705771516338,
+    "producer_headroom": 7.056713269461715,
     "sustainable": true,
-    "buffer_seconds": 60.38336770126681,
+    "buffer_seconds": 60.76819364240618,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356761.7305578408,
-    "upload_bytes_per_second": 4544700.995911101,
+    "generation_bytes_per_second": 355279.84770351776,
+    "upload_bytes_per_second": 4548455.371141299,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.5527623896734,
+    "estimated_full_shards_per_hour": 76.234784825603,
     "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": 1.0596165809984086,
-      "bytes": 5348616,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3328053339973849,
       "bytes": 7544008,
@@ -562,6 +556,12 @@
       "bytes": 5714275,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.098721825001121,
+      "bytes": 5875494,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 185700, "loss": 0.6390447355806828, "lr": 0.0003, "grad_norm": 0.17158830601594321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27178325663, "tokens_per_second": 150389.95777772815, "data_wait_s": 8.137001714203507e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185800, "loss": 0.5502657704055309, "lr": 0.0003, "grad_norm": 0.14309495935863623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27192935771, "tokens_per_second": 149516.55377683364, "data_wait_s": 1.3199998647905886e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 185900, "loss": 0.6017749421298504, "lr": 0.0003, "grad_norm": 0.13863636085678446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27207593487, "tokens_per_second": 150194.8504445067, "data_wait_s": 7.1880022005643696e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 186000, "loss": 0.5967893414199352, "lr": 0.0003, "grad_norm": 0.15444484948419754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27222196967, "tokens_per_second": 149711.13759823842, "data_wait_s": 1.0324001777917147e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fc9652fd2370412885cd05bc142d5c374e062556b24c3f9cea3aec2a43139fe
-size 659325
+oid sha256:2359b76472c0e33795cfcef222a80c88f2e4ccca47bc0a3dc72da5d1c7db6944
+size 661443