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": 75556.2576872194,
-      "required_pages_per_second_upper_estimate": 0.006549024920743807
+      "required_token_positions_per_second": 75200.3601959549,
+      "required_pages_per_second_upper_estimate": 0.006518176628214945
     },
     {
       "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": 9444.532210902426,
-      "required_pages_per_second_upper_estimate": 0.003767565107269198
+      "required_token_positions_per_second": 9400.045024494362,
+      "required_pages_per_second_upper_estimate": 0.003749818503468311
     },
     {
       "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": 18889.06442180485,
-      "required_pages_per_second_upper_estimate": 0.004369652617804834
+      "required_token_positions_per_second": 18800.090048988724,
+      "required_pages_per_second_upper_estimate": 0.004349069962549303
     },
     {
       "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": 28333.59663270728,
-      "required_pages_per_second_upper_estimate": 0.0017253673800389968
+      "required_token_positions_per_second": 28200.13507348309,
+      "required_pages_per_second_upper_estimate": 0.0017172402713009293
     },
     {
       "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": 28333.59663270728,
-      "required_pages_per_second_upper_estimate": 0.0021811987012011836
+      "required_token_positions_per_second": 28200.13507348309,
+      "required_pages_per_second_upper_estimate": 0.002170924460926864
     },
     {
       "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": 28333.59663270728,
-      "required_pages_per_second_upper_estimate": 0.35180844373030135
+      "required_token_positions_per_second": 28200.13507348309,
+      "required_pages_per_second_upper_estimate": 0.35015129783184235
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10101006653127342,
-    "producer_p95_seconds": 0.280451587997959,
-    "gpu_update_seconds": 0.975802696650384,
-    "producer_batches_per_second": 9.900003379271045,
-    "gpu_batches_per_second": 1.0247973319121555,
-    "producer_headroom": 9.6604499943406,
+    "producer_batch_seconds": 0.13479405474211603,
+    "producer_p95_seconds": 0.3213546050028526,
+    "gpu_update_seconds": 0.9804208358561279,
+    "producer_batches_per_second": 7.418724823681357,
+    "gpu_batches_per_second": 1.0199701632480862,
+    "producer_headroom": 7.273472392620282,
     "sustainable": true,
-    "buffer_seconds": 60.49976719232381,
+    "buffer_seconds": 60.78609182307993,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424260.16566595936,
-    "upload_bytes_per_second": 4035375.0496331183,
+    "generation_bytes_per_second": 402796.8360344785,
+    "upload_bytes_per_second": 4084200.961932447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.0363552807244,
+    "estimated_full_shards_per_hour": 86.43082438255087,
     "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": 0.827956348999578,
-      "bytes": 1165763,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.215623644999141,
       "bytes": 7281440,
@@ -562,6 +556,12 @@
       "bytes": 955823,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.141221003003011,
+      "bytes": 7503350,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -268,3 +268,4 @@
 {"step": 341800, "loss": 0.5851886924356222, "lr": 0.0003, "grad_norm": 0.1859805109652498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50030654863, "tokens_per_second": 149067.29383108535, "data_wait_s": 1.384899951517582e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341900, "loss": 0.66478824429214, "lr": 0.0003, "grad_norm": 0.17863731514469988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50045282204, "tokens_per_second": 148654.6301315183, "data_wait_s": 1.461700230720453e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342000, "loss": 0.5790649708360434, "lr": 0.0003, "grad_norm": 0.1369144824027103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50059709616, "tokens_per_second": 145901.03387296785, "data_wait_s": 1.6900001355679706e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 342100, "loss": 0.5834092274308205, "lr": 0.0003, "grad_norm": 0.1502677618109866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50074220963, "tokens_per_second": 148094.98289689445, "data_wait_s": 1.0193001799052581e-05, "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 sha256:2cd23eb7ce8ed53ad44d5dc94749bc3b4e992807623348a0e073047e6e2e56ba
-size 575940
+oid sha256:eaa7fdcb69e59b5e900551aab69de8c45b3291058f8b551c00e6fc104755c427
+size 578058