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": 76796.20367691354,
-      "required_pages_per_second_upper_estimate": 0.005312939562111441
+      "required_token_positions_per_second": 76826.11512206601,
+      "required_pages_per_second_upper_estimate": 0.005315008905291203
     },
     {
       "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": 9599.525459614193,
-      "required_pages_per_second_upper_estimate": 0.004019120849053698
+      "required_token_positions_per_second": 9603.264390258251,
+      "required_pages_per_second_upper_estimate": 0.0040206862612366156
     },
     {
       "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": 19199.050919228386,
-      "required_pages_per_second_upper_estimate": 0.003998655993454457
+      "required_token_positions_per_second": 19206.528780516503,
+      "required_pages_per_second_upper_estimate": 0.004000213434756308
     },
     {
       "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": 28798.576378842583,
-      "required_pages_per_second_upper_estimate": 0.0017264306810315948
+      "required_token_positions_per_second": 28809.793170774756,
+      "required_pages_per_second_upper_estimate": 0.0017271031105808789
     },
     {
       "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": 28798.576378842583,
-      "required_pages_per_second_upper_estimate": 0.0022389626419045826
+      "required_token_positions_per_second": 28809.793170774756,
+      "required_pages_per_second_upper_estimate": 0.0022398346981398554
     },
     {
       "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": 28798.576378842583,
-      "required_pages_per_second_upper_estimate": 0.36939721628561184
+      "required_token_positions_per_second": 28809.793170774756,
+      "required_pages_per_second_upper_estimate": 0.3695410932488649
     }
   ],
   "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": 0.07670450882821456,
-    "producer_p95_seconds": 0.2733304360008333,
-    "gpu_update_seconds": 0.9600474563844109,
-    "producer_batches_per_second": 13.037043262209973,
-    "gpu_batches_per_second": 1.0416151757393874,
-    "producer_headroom": 12.516180222658207,
+    "producer_batch_seconds": 0.0950199218204375,
+    "producer_p95_seconds": 0.2645207030000165,
+    "gpu_update_seconds": 0.9596736719389815,
+    "producer_batches_per_second": 10.524108848350089,
+    "gpu_batches_per_second": 1.0420208756790637,
+    "producer_headroom": 10.099710182381655,
     "sustainable": true,
-    "buffer_seconds": 60.48298975221789,
+    "buffer_seconds": 60.45944133215583,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307265.3843205351,
-    "upload_bytes_per_second": 4962259.811977918,
+    "generation_bytes_per_second": 333176.24756099796,
+    "upload_bytes_per_second": 4853699.577199489,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.93199870311776,
+    "estimated_full_shards_per_hour": 71.4918667804952,
     "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": 1.3531785280028998,
-      "bytes": 11787584,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.0967427190007584,
       "bytes": 6096101,
@@ -562,6 +556,12 @@
       "bytes": 12113021,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.0979482909970102,
+      "bytes": 1151793,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 867100, "loss": 0.5442878603935242, "lr": 3e-05, "grad_norm": 0.31072423824450185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126919416590, "tokens_per_second": 152392.37245601066, "data_wait_s": 1.6568999853916466e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 867200, "loss": 0.4785754680633545, "lr": 3e-05, "grad_norm": 0.2828800902508147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126934086501, "tokens_per_second": 151412.1845326729, "data_wait_s": 1.2123997294111177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 867300, "loss": 0.500617504119873, "lr": 3e-05, "grad_norm": 0.3123256527886336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126948696202, "tokens_per_second": 151520.75081222452, "data_wait_s": 1.4877998182782903e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 867400, "loss": 0.5173141360282898, "lr": 3e-05, "grad_norm": 0.29034363173134775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126963374771, "tokens_per_second": 151831.47326916485, "data_wait_s": 8.228002116084099e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e349fd3944bfdd85558117e4aca0ad0bd13f7f1bbe7cc52731df889590b3fc1
-size 529344
+oid sha256:91a0495f82959d68cc81665d845924cc2fb3e832c8020adc13d20971ba07cfcc
+size 531462