Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76666.71321686973,
-      "required_pages_per_second_upper_estimate": 0.005303981111626347
+      "required_token_positions_per_second": 77065.46180712563,
+      "required_pages_per_second_upper_estimate": 0.00533156746432445
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9583.339152108716,
-      "required_pages_per_second_upper_estimate": 0.004012343980109692
+      "required_token_positions_per_second": 9633.182725890703,
+      "required_pages_per_second_upper_estimate": 0.00403321244360003
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19166.678304217432,
-      "required_pages_per_second_upper_estimate": 0.003991913631470943
+      "required_token_positions_per_second": 19266.365451781407,
+      "required_pages_per_second_upper_estimate": 0.00401267583538165
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28750.01745632615,
-      "required_pages_per_second_upper_estimate": 0.0017235196477719161
+      "required_token_positions_per_second": 28899.54817767211,
+      "required_pages_per_second_upper_estimate": 0.001732483786196415
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28750.01745632615,
-      "required_pages_per_second_upper_estimate": 0.0022351873992670574
+      "required_token_positions_per_second": 28899.54817767211,
+      "required_pages_per_second_upper_estimate": 0.002246812755135572
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28750.01745632615,
-      "required_pages_per_second_upper_estimate": 0.3687743545660798
+      "required_token_positions_per_second": 28899.54817767211,
+      "required_pages_per_second_upper_estimate": 0.37069237410592615
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14815534935127062,
-    "producer_p95_seconds": 0.3474937040009536,
-    "gpu_update_seconds": 0.9616689813145257,
-    "producer_batches_per_second": 6.749671911130516,
-    "gpu_batches_per_second": 1.03985884897013,
-    "producer_headroom": 6.490950110984151,
+    "producer_batch_seconds": 0.2583873071171183,
+    "producer_p95_seconds": 0.3262074960002792,
+    "gpu_update_seconds": 0.95669315762386,
+    "producer_batches_per_second": 3.87015914658197,
+    "gpu_batches_per_second": 1.0452672228614044,
+    "producer_headroom": 3.702554774450368,
     "sustainable": true,
-    "buffer_seconds": 60.58514582281512,
+    "buffer_seconds": 60.27166893030318,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352942.01542088424,
-    "upload_bytes_per_second": 4648628.037487278,
+    "generation_bytes_per_second": 352217.4872777173,
+    "upload_bytes_per_second": 4624970.714440824,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.73314043970008,
+    "estimated_full_shards_per_hour": 75.57767356632841,
     "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": 1.3205864709998423,
-      "bytes": 10433001,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.281639933000406,
       "bytes": 7384470,
@@ -562,6 +556,12 @@
       "bytes": 10319080,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3421651660028147,
+      "bytes": 8322765,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -392,3 +392,4 @@
 {"step": 799200, "loss": 0.5641132593154907, "lr": 3e-05, "grad_norm": 0.28709438999567377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116979520606, "tokens_per_second": 137407.55568114898, "data_wait_s": 1.556699862703681e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799300, "loss": 0.5825047492980957, "lr": 3e-05, "grad_norm": 0.28232000321448114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116994211030, "tokens_per_second": 152123.7532338152, "data_wait_s": 2.0558996766339988e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799400, "loss": 0.5226407647132874, "lr": 3e-05, "grad_norm": 0.3258039355682198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117008882450, "tokens_per_second": 152703.17333596997, "data_wait_s": 1.1236996215302497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 799500, "loss": 0.5818156003952026, "lr": 3e-05, "grad_norm": 0.26518255366618954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117023520607, "tokens_per_second": 152479.0557962953, "data_wait_s": 1.111400342779234e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94b67d11a6dd0bcde03aad1832f5a767956f4f6bc8daba54f572bc21a2f20b44
-size 840138
+oid sha256:eeb0061171e3a82d1e57b4689af45a09f5eab399e7313aa86c2458894a46b03e
+size 842256