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": 75420.91002438392,
-      "required_pages_per_second_upper_estimate": 0.006537293328364743
+      "required_token_positions_per_second": 75695.91670086801,
+      "required_pages_per_second_upper_estimate": 0.0065611302100843346
     },
     {
       "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": 9427.61375304799,
-      "required_pages_per_second_upper_estimate": 0.003760816081477577
+      "required_token_positions_per_second": 9461.989587608501,
+      "required_pages_per_second_upper_estimate": 0.003774529115848293
     },
     {
       "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": 18855.22750609598,
-      "required_pages_per_second_upper_estimate": 0.004361825042864964
+      "required_token_positions_per_second": 18923.979175217002,
+      "required_pages_per_second_upper_estimate": 0.004377729531528062
     },
     {
       "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": 28282.84125914397,
-      "required_pages_per_second_upper_estimate": 0.0017222766440818562
+      "required_token_positions_per_second": 28385.968762825505,
+      "required_pages_per_second_upper_estimate": 0.001728556567987866
     },
     {
       "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": 28282.84125914397,
-      "required_pages_per_second_upper_estimate": 0.002177291412044414
+      "required_token_positions_per_second": 28385.968762825505,
+      "required_pages_per_second_upper_estimate": 0.0021852304527530198
     },
     {
       "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": 28282.84125914397,
-      "required_pages_per_second_upper_estimate": 0.3511782318579531
+      "required_token_positions_per_second": 28385.968762825505,
+      "required_pages_per_second_upper_estimate": 0.3524587303081255
     }
   ],
   "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": 0.09297166586765115,
-    "producer_p95_seconds": 0.3729305309971096,
-    "gpu_update_seconds": 0.97755383720726,
-    "producer_batches_per_second": 10.755965171405443,
-    "gpu_batches_per_second": 1.0229615617456587,
-    "producer_headroom": 10.514535026175034,
+    "producer_batch_seconds": 0.10955438575786047,
+    "producer_p95_seconds": 0.3829696110042278,
+    "gpu_update_seconds": 0.9740023400648579,
+    "producer_batches_per_second": 9.127886511181963,
+    "gpu_batches_per_second": 1.0266915785165476,
+    "producer_headroom": 8.890582821737684,
     "sustainable": true,
-    "buffer_seconds": 60.60833790685012,
+    "buffer_seconds": 60.38814508402119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491254.8961145878,
-    "upload_bytes_per_second": 4024317.616997542,
+    "generation_bytes_per_second": 491021.5619626832,
+    "upload_bytes_per_second": 4664112.175691686,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.41186487749314,
+    "estimated_full_shards_per_hour": 105.36179680023548,
     "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": 15.209589787002187,
-      "bytes": 6010603,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.505151831999683,
       "bytes": 12552564,
@@ -562,6 +556,12 @@
       "bytes": 7522789,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.7894636139972135,
+      "bytes": 5936931,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -330,3 +330,4 @@
 {"step": 73000, "loss": 0.6199126038700342, "lr": 0.0003, "grad_norm": 0.12562120347064035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10683593832, "tokens_per_second": 149884.9240272704, "data_wait_s": 8.317998435813934e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73100, "loss": 0.6249210871756077, "lr": 0.0003, "grad_norm": 0.13517386636139803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10698270266, "tokens_per_second": 149889.29134254056, "data_wait_s": 8.770999556872994e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73200, "loss": 0.6195316538214684, "lr": 0.0003, "grad_norm": 0.15515854729561093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10712902577, "tokens_per_second": 149556.26445771017, "data_wait_s": 7.280999852810055e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 73300, "loss": 0.6037278231233358, "lr": 0.0003, "grad_norm": 0.13094053326599112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10727361084, "tokens_per_second": 146893.6334221823, "data_wait_s": 1.1030999303329736e-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:edbab4166d5981708453e8e8a941cc6a7b85ebec0ecb0e8cf8ccf5d9636c1d89
-size 708039
+oid sha256:8c546b801db4774f082755b1174af2004c8597e37234d6b49fa2bec0d5c01fbe
+size 710157