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": 76929.7203127848,
-      "required_pages_per_second_upper_estimate": 0.005322176552782292
+      "required_token_positions_per_second": 76877.34127819644,
+      "required_pages_per_second_upper_estimate": 0.005318552849633364
     },
     {
       "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": 9616.2150390981,
-      "required_pages_per_second_upper_estimate": 0.00402610842746556
+      "required_token_positions_per_second": 9609.667659774555,
+      "required_pages_per_second_upper_estimate": 0.004023367176467619
     },
     {
       "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": 19232.4300781962,
-      "required_pages_per_second_upper_estimate": 0.004005607991999835
+      "required_token_positions_per_second": 19219.33531954911,
+      "required_pages_per_second_upper_estimate": 0.004002880699105655
     },
     {
       "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": 28848.6451172943,
-      "required_pages_per_second_upper_estimate": 0.0017294322254512383
+      "required_token_positions_per_second": 28829.002979323664,
+      "required_pages_per_second_upper_estimate": 0.0017282547092717067
     },
     {
       "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": 28848.6451172943,
-      "required_pages_per_second_upper_estimate": 0.0022428552660900976
+      "required_token_positions_per_second": 28829.002979323664,
+      "required_pages_per_second_upper_estimate": 0.0022413281762595147
     },
     {
       "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": 28848.6451172943,
-      "required_pages_per_second_upper_estimate": 0.3700394443028476
+      "required_token_positions_per_second": 28829.002979323664,
+      "required_pages_per_second_upper_estimate": 0.36978749604704486
     }
   ],
   "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": 0.3709434043204567,
-    "producer_p95_seconds": 0.24223145600262797,
-    "gpu_update_seconds": 0.9583812302999791,
-    "producer_batches_per_second": 2.69582903578494,
-    "gpu_batches_per_second": 1.0434261110132488,
-    "producer_headroom": 2.5836319479939776,
+    "producer_batch_seconds": 0.09012341921871325,
+    "producer_p95_seconds": 0.250304089000565,
+    "gpu_update_seconds": 0.9590342066227303,
+    "producer_batches_per_second": 11.095895036707171,
+    "gpu_batches_per_second": 1.042715674888732,
+    "producer_headroom": 10.641342893297553,
     "sustainable": true,
-    "buffer_seconds": 60.37801750889869,
+    "buffer_seconds": 60.41915501723201,
     "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": 313762.09127828694,
-    "upload_bytes_per_second": 1904513.2746880879,
+    "generation_bytes_per_second": 313015.03246302105,
+    "upload_bytes_per_second": 1922604.6302654655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.32604078065354,
+    "estimated_full_shards_per_hour": 67.16573934953665,
     "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": 1.055462077001721,
-      "bytes": 6384191,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3574712270001328,
       "bytes": 12582265,
@@ -562,6 +556,12 @@
       "bytes": 7563007,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2517445409976062,
+      "bytes": 10940783,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 665400, "loss": 0.5959066152572632, "lr": 0.0001, "grad_norm": 0.1950656503484615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97397977138, "tokens_per_second": 154085.35871400713, "data_wait_s": 7.749000360490754e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 665500, "loss": 0.5516788959503174, "lr": 0.0001, "grad_norm": 0.19118081544389923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97412679412, "tokens_per_second": 152933.01792986688, "data_wait_s": 1.3611999747809023e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 665600, "loss": 0.5553975701332092, "lr": 0.0001, "grad_norm": 0.20037151667809544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97427422991, "tokens_per_second": 152960.84991280807, "data_wait_s": 1.804899875423871e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 665700, "loss": 0.5386643409729004, "lr": 0.0001, "grad_norm": 0.17973047732928352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97442145075, "tokens_per_second": 153621.0965668127, "data_wait_s": 1.4173998351907358e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93ea643d5f4eac247518fd50c14147a70744f292bf8e481307cfe36e607e7514
-size 439605
+oid sha256:2686e43b22b57fa8f743ce6930fde5452a9dab0b642c428b32292c7b6093e48f
+size 441723