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": 76689.59081638904,
-      "required_pages_per_second_upper_estimate": 0.005305563837044437
+      "required_token_positions_per_second": 77040.69559298594,
+      "required_pages_per_second_upper_estimate": 0.005329854080164733
     },
     {
       "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": 9586.19885204863,
-      "required_pages_per_second_upper_estimate": 0.004013541276757209
+      "required_token_positions_per_second": 9630.086949123242,
+      "required_pages_per_second_upper_estimate": 0.004031916306514665
     },
     {
       "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": 19172.39770409726,
-      "required_pages_per_second_upper_estimate": 0.003993104831635194
+      "required_token_positions_per_second": 19260.173898246485,
+      "required_pages_per_second_upper_estimate": 0.004011386298062573
     },
     {
       "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": 28758.59655614589,
-      "required_pages_per_second_upper_estimate": 0.0017240339517065825
+      "required_token_positions_per_second": 28890.26084736973,
+      "required_pages_per_second_upper_estimate": 0.0017319270249257196
     },
     {
       "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": 28758.59655614589,
-      "required_pages_per_second_upper_estimate": 0.00223585438654257
+      "required_token_positions_per_second": 28890.26084736973,
+      "required_pages_per_second_upper_estimate": 0.0022460907060552105
     },
     {
       "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": 28758.59655614589,
-      "required_pages_per_second_upper_estimate": 0.36888439804704776
+      "required_token_positions_per_second": 28890.26084736973,
+      "required_pages_per_second_upper_estimate": 0.37057324620476556
     }
   ],
   "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": 0.30072454796845705,
-    "producer_p95_seconds": 0.3476668540024548,
-    "gpu_update_seconds": 0.9613821017316455,
-    "producer_batches_per_second": 3.325302196829272,
-    "gpu_batches_per_second": 1.040169146272638,
-    "producer_headroom": 3.1968860148805835,
+    "producer_batch_seconds": 0.1692697002345085,
+    "producer_p95_seconds": 0.3418203010005527,
+    "gpu_update_seconds": 0.9570007050496109,
+    "producer_batches_per_second": 5.9077318540446795,
+    "gpu_batches_per_second": 1.0449313095836852,
+    "producer_headroom": 5.653703549564804,
     "sustainable": true,
-    "buffer_seconds": 60.56707240909367,
+    "buffer_seconds": 60.29104441812549,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328587.85388796235,
-    "upload_bytes_per_second": 4490678.957494245,
+    "generation_bytes_per_second": 336800.4776230626,
+    "upload_bytes_per_second": 4544156.954977472,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.50730431059984,
+    "estimated_full_shards_per_hour": 72.26954218405636,
     "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": 1.3202570079956786,
-      "bytes": 6097373,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.580753474998346,
       "bytes": 10855257,
@@ -562,6 +556,12 @@
       "bytes": 6309563,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.7539230830007,
+      "bytes": 13173822,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 801500, "loss": 0.5405556559562683, "lr": 3e-05, "grad_norm": 0.25965025620707455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117316172213, "tokens_per_second": 151488.74701542806, "data_wait_s": 1.4527999155689031e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801600, "loss": 0.5029669404029846, "lr": 3e-05, "grad_norm": 0.28534698327365837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117330622104, "tokens_per_second": 149872.91781004344, "data_wait_s": 1.5680998330935836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801700, "loss": 0.5376384854316711, "lr": 3e-05, "grad_norm": 0.25954194731941643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117345270836, "tokens_per_second": 151934.27009175546, "data_wait_s": 1.455999881727621e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 801800, "loss": 0.505940318107605, "lr": 3e-05, "grad_norm": 0.27344645578183735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117359953415, "tokens_per_second": 153007.89898270488, "data_wait_s": 9.234994649887085e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c866ccdeb0d2819ae51e32232e9a7f770e1d851e8f2ffeaf4a63ed47a7d41b74
-size 889635
+oid sha256:4706f3cea72f64f60c2d316e5a6a146f47daa6f9eedfcd00e749644837f2eeb7
+size 891753