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": 74662.74045688349,
-      "required_pages_per_second_upper_estimate": 0.006471577111817007
+      "required_token_positions_per_second": 74511.58170452456,
+      "required_pages_per_second_upper_estimate": 0.006458475054270891
     },
     {
       "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": 9332.842557110436,
-      "required_pages_per_second_upper_estimate": 0.0037230104344624366
+      "required_token_positions_per_second": 9313.94771306557,
+      "required_pages_per_second_upper_estimate": 0.003715472998669846
     },
     {
       "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": 18665.685114220872,
-      "required_pages_per_second_upper_estimate": 0.004317977746336822
+      "required_token_positions_per_second": 18627.89542613114,
+      "required_pages_per_second_upper_estimate": 0.004309235767073058
     },
     {
       "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": 27998.52767133131,
-      "required_pages_per_second_upper_estimate": 0.0017049634382621777
+      "required_token_positions_per_second": 27941.843139196713,
+      "required_pages_per_second_upper_estimate": 0.0017015116476559364
     },
     {
       "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": 27998.52767133131,
-      "required_pages_per_second_upper_estimate": 0.002155404165024212
+      "required_token_positions_per_second": 27941.843139196713,
+      "required_pages_per_second_upper_estimate": 0.002151040432827663
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27998.52767133131,
-      "required_pages_per_second_upper_estimate": 0.3476480086336878
+      "required_token_positions_per_second": 27941.843139196713,
+      "required_pages_per_second_upper_estimate": 0.34694417645550135
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13071090790640483,
-    "producer_p95_seconds": 0.4467034029948991,
-    "gpu_update_seconds": 0.9874804962801588,
-    "producer_batches_per_second": 7.6504709210347395,
-    "gpu_batches_per_second": 1.012678228853129,
-    "producer_headroom": 7.554690821880309,
+    "producer_batch_seconds": 0.12054120989080275,
+    "producer_p95_seconds": 0.4255806700020912,
+    "gpu_update_seconds": 0.9894837596169699,
+    "producer_batches_per_second": 8.29591805910934,
+    "gpu_batches_per_second": 1.0106280070600664,
+    "producer_headroom": 8.208676190601826,
     "sustainable": true,
-    "buffer_seconds": 60.236310273089686,
+    "buffer_seconds": 60.358509336635166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352938.3446217756,
-    "upload_bytes_per_second": 3665479.853821214,
+    "generation_bytes_per_second": 359104.02409992594,
+    "upload_bytes_per_second": 5053663.424818994,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.7323527716632,
+    "estimated_full_shards_per_hour": 77.05536405800184,
     "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": 33.798858237001696,
-      "bytes": 12238534,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.911944024002878,
       "bytes": 5136667,
@@ -562,6 +556,12 @@
       "bytes": 1123327,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1691736089996994,
+      "bytes": 7423568,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -413,3 +413,4 @@
 {"step": 41300, "loss": 0.6393550038337708, "lr": 0.0003, "grad_norm": 0.1304140443170049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6043338175, "tokens_per_second": 146804.45269256743, "data_wait_s": 1.0766001651063561e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41400, "loss": 0.6243662759661674, "lr": 0.0003, "grad_norm": 0.1266594176148852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6057938636, "tokens_per_second": 147845.282010544, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41500, "loss": 0.6293739005923271, "lr": 0.0003, "grad_norm": 0.13255592951291514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6072666804, "tokens_per_second": 148697.67904327752, "data_wait_s": 1.361100294161588e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 41600, "loss": 0.6368713565170765, "lr": 0.0003, "grad_norm": 0.15394438290790816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6087361539, "tokens_per_second": 148919.23862411396, "data_wait_s": 1.1239004379604012e-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:c6c592769cb04e9535b0f94aaa09398025e61e61b6adf06d9d2b77dcf4229096
-size 879625
+oid sha256:489ceeb471dad4ef0b05021bdaf3878674acd0139de0278a897a1a3781872280
+size 881743