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": 74694.75588854252,
-      "required_pages_per_second_upper_estimate": 0.006474352128290842
+      "required_token_positions_per_second": 74774.2771622577,
+      "required_pages_per_second_upper_estimate": 0.006481244830751648
     },
     {
       "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": 9336.844486067816,
-      "required_pages_per_second_upper_estimate": 0.003724606863757705
+      "required_token_positions_per_second": 9346.784645282212,
+      "required_pages_per_second_upper_estimate": 0.0037285721418869526
     },
     {
       "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": 18673.68897213563,
-      "required_pages_per_second_upper_estimate": 0.004319829298002313
+      "required_token_positions_per_second": 18693.569290564425,
+      "required_pages_per_second_upper_estimate": 0.00432442825978916
     },
     {
       "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": 28010.53345820345,
-      "required_pages_per_second_upper_estimate": 0.0017056945276931951
+      "required_token_positions_per_second": 28040.353935846633,
+      "required_pages_per_second_upper_estimate": 0.0017075104383257106
     },
     {
       "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": 28010.53345820345,
-      "required_pages_per_second_upper_estimate": 0.002156328403731774
+      "required_token_positions_per_second": 28040.353935846633,
+      "required_pages_per_second_upper_estimate": 0.002158624066649112
     },
     {
       "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": 28010.53345820345,
-      "required_pages_per_second_upper_estimate": 0.3477970803258558
+      "required_token_positions_per_second": 28040.353935846633,
+      "required_pages_per_second_upper_estimate": 0.34816735085546563
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.20007373156266794,
-    "producer_p95_seconds": 0.37525636500140536,
-    "gpu_update_seconds": 0.9870572454914359,
-    "producer_batches_per_second": 4.998157390225791,
-    "gpu_batches_per_second": 1.0131124659361779,
-    "producer_headroom": 4.933467466128934,
+    "producer_batch_seconds": 0.11320975642215103,
+    "producer_p95_seconds": 0.3445786880001833,
+    "gpu_update_seconds": 0.9860075255560504,
+    "producer_batches_per_second": 8.833160953646715,
+    "gpu_batches_per_second": 1.0141910422398233,
+    "producer_headroom": 8.70956317474352,
     "sustainable": true,
-    "buffer_seconds": 60.21049197497759,
+    "buffer_seconds": 60.14645905891908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325832.25086374395,
-    "upload_bytes_per_second": 4211750.236913444,
+    "generation_bytes_per_second": 314167.36518101394,
+    "upload_bytes_per_second": 4237055.285045125,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.91601604875792,
+    "estimated_full_shards_per_hour": 67.41300312588514,
     "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": 1.035739069000556,
-      "bytes": 2751840,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1663796350003395,
       "bytes": 6310887,
@@ -562,6 +556,12 @@
       "bytes": 12663721,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.2500024359978852,
+      "bytes": 6221044,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 345200, "loss": 0.5913329981267452, "lr": 0.0003, "grad_norm": 0.14752144068889433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50527916156, "tokens_per_second": 148072.05477990818, "data_wait_s": 1.4799999917158857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345300, "loss": 0.6465697512030602, "lr": 0.0003, "grad_norm": 0.17317448365114063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50542655645, "tokens_per_second": 149837.02061501687, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345400, "loss": 0.5820652358233929, "lr": 0.0003, "grad_norm": 0.15884571989075166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50557407659, "tokens_per_second": 149447.22265391704, "data_wait_s": 8.90400042408146e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 345500, "loss": 0.5766577012836933, "lr": 0.0003, "grad_norm": 0.1543777721993438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50572044161, "tokens_per_second": 148827.2971557669, "data_wait_s": 7.553000614279881e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d22e33c8b970dc325de5ab17ed12fd917c8f926acd0fb25d55d82e9ff126b909
-size 648735
+oid sha256:16d1d076a2d5cb9ad10581e559049a5fab3e0edc68ef215dc454f669823b6fc7
+size 650853