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": 75108.95336260133,
-      "required_pages_per_second_upper_estimate": 0.006510253715568359
+      "required_token_positions_per_second": 75397.03422616019,
+      "required_pages_per_second_upper_estimate": 0.0065352238346873676
     },
     {
       "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": 9388.619170325166,
-      "required_pages_per_second_upper_estimate": 0.0037452605594084755
+      "required_token_positions_per_second": 9424.629278270024,
+      "required_pages_per_second_upper_estimate": 0.003759625529866772
     },
     {
       "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": 18777.238340650332,
-      "required_pages_per_second_upper_estimate": 0.004343783622001546
+      "required_token_positions_per_second": 18849.258556540048,
+      "required_pages_per_second_upper_estimate": 0.0043604442314351234
     },
     {
       "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": 28165.8575109755,
-      "required_pages_per_second_upper_estimate": 0.0017151529475846888
+      "required_token_positions_per_second": 28273.88783481007,
+      "required_pages_per_second_upper_estimate": 0.0017217314275149089
     },
     {
       "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": 28165.8575109755,
-      "required_pages_per_second_upper_estimate": 0.002168285679278664
+      "required_token_positions_per_second": 28273.88783481007,
+      "required_pages_per_second_upper_estimate": 0.002176602152654525
     },
     {
       "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": 28165.8575109755,
-      "required_pages_per_second_upper_estimate": 0.34972568522512015
+      "required_token_positions_per_second": 28273.88783481007,
+      "required_pages_per_second_upper_estimate": 0.35106706029290974
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05069276689844315,
-    "producer_p95_seconds": 0.17191456600085075,
-    "gpu_update_seconds": 0.981613998055138,
-    "producer_batches_per_second": 19.726680179114695,
-    "gpu_batches_per_second": 1.018730378724519,
-    "producer_headroom": 19.36398539897582,
+    "producer_batch_seconds": 0.0677809657108881,
+    "producer_p95_seconds": 0.18355977700048243,
+    "gpu_update_seconds": 0.9778633968392739,
+    "producer_batches_per_second": 14.753404433117474,
+    "gpu_batches_per_second": 1.02263772550673,
+    "producer_headroom": 14.426814173911856,
     "sustainable": true,
-    "buffer_seconds": 60.86006787941856,
+    "buffer_seconds": 60.62753060403498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440281.4769263359,
-    "upload_bytes_per_second": 6126865.666387042,
+    "generation_bytes_per_second": 462575.1781962507,
+    "upload_bytes_per_second": 5726288.23889271,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.47415572016294,
+    "estimated_full_shards_per_hour": 99.25786504188194,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 10970126,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.999679897000533,
+      "bytes": 1109612,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 43800, "loss": 0.6106060147285461, "lr": 0.0003, "grad_norm": 0.13517690915616656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6409946152, "tokens_per_second": 147930.89778770387, "data_wait_s": 1.0691999705159105e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 43900, "loss": 0.603788185864687, "lr": 0.0003, "grad_norm": 0.1256535827192673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6424410562, "tokens_per_second": 147235.09427609958, "data_wait_s": 1.493499985372182e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 44000, "loss": 0.608014291152358, "lr": 0.0003, "grad_norm": 0.1289988112922647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6438836210, "tokens_per_second": 146621.41609845211, "data_wait_s": 1.3150000086170621e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 44100, "loss": 0.6197555270045996, "lr": 0.0003, "grad_norm": 0.13371724287142478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6453322297, "tokens_per_second": 147163.8639494913, "data_wait_s": 1.3975999536341988e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74bcc030d2b9e78be37e6b4c499a09529cb39fcbaa904d9d3311cdafeab9ef87
-size 84885
+oid sha256:9485df45e1b0dbd84aa857f5d1d536af8f0026478a165e40f80de87b9592f7df
+size 87003