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": 77389.30771722819,
-      "required_pages_per_second_upper_estimate": 0.005353971875811383
+      "required_token_positions_per_second": 76510.47617266774,
+      "required_pages_per_second_upper_estimate": 0.005293172270388565
     },
     {
       "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": 9673.663464653524,
-      "required_pages_per_second_upper_estimate": 0.004050160883586072
+      "required_token_positions_per_second": 9563.809521583467,
+      "required_pages_per_second_upper_estimate": 0.004004167331633831
     },
     {
       "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": 19347.326929307048,
-      "required_pages_per_second_upper_estimate": 0.0040295379760525
+      "required_token_positions_per_second": 19127.619043166935,
+      "required_pages_per_second_upper_estimate": 0.003983778617456107
     },
     {
       "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": 29020.99039396057,
-      "required_pages_per_second_upper_estimate": 0.0017397640616313808
+      "required_token_positions_per_second": 28691.428564750404,
+      "required_pages_per_second_upper_estimate": 0.0017200073331820076
     },
     {
       "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": 29020.99039396057,
-      "required_pages_per_second_upper_estimate": 0.002256254353284143
+      "required_token_positions_per_second": 28691.428564750404,
+      "required_pages_per_second_upper_estimate": 0.002230632370652342
     },
     {
       "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": 29020.99039396057,
-      "required_pages_per_second_upper_estimate": 0.37225010446198187
+      "required_token_positions_per_second": 28691.428564750404,
+      "required_pages_per_second_upper_estimate": 0.3680228391727967
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2763804745312157,
-    "producer_p95_seconds": 0.18456472699926962,
-    "gpu_update_seconds": 0.9526897471339815,
-    "producer_batches_per_second": 3.618200604424591,
-    "gpu_batches_per_second": 1.0496596641334117,
-    "producer_headroom": 3.4470226189092825,
+    "producer_batch_seconds": 0.42690014150778666,
+    "producer_p95_seconds": 0.17914704600025289,
+    "gpu_update_seconds": 0.9636327427058708,
+    "producer_batches_per_second": 2.342468185810522,
+    "gpu_batches_per_second": 1.0377397484357063,
+    "producer_headroom": 2.2572790425938387,
     "sustainable": true,
-    "buffer_seconds": 60.01945406944083,
+    "buffer_seconds": 60.708862790469865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383627.1420612841,
-    "upload_bytes_per_second": 2654223.418381419,
+    "generation_bytes_per_second": 358172.58723321615,
+    "upload_bytes_per_second": 2806400.2323788204,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.31745430354016,
+    "estimated_full_shards_per_hour": 76.85549938914646,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 5978754,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5570587419997537,
+      "bytes": 16077242,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -71,3 +71,4 @@
 {"step": 652100, "loss": 0.5375281572341919, "lr": 0.0001, "grad_norm": 0.1803161872897369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95449862098, "tokens_per_second": 151957.2287803522, "data_wait_s": 1.1095999980170745e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 652200, "loss": 0.5419741272926331, "lr": 0.0001, "grad_norm": 0.1745661499664961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95464388264, "tokens_per_second": 151707.58956482777, "data_wait_s": 2.0442999812075868e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 652300, "loss": 0.5161867737770081, "lr": 0.0001, "grad_norm": 0.19488258772910172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95478960178, "tokens_per_second": 151078.89930506705, "data_wait_s": 7.265000022016466e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 652400, "loss": 0.5412026047706604, "lr": 0.0001, "grad_norm": 0.1897618164832828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95493556802, "tokens_per_second": 152952.9116296233, "data_wait_s": 1.805299962143181e-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 sha256:2ca0c386d006f8b75bcc07722a228ae5db21a56abbb181f6e1170ab0cd5792e5
-size 155562
+oid sha256:1116c572c8cf51b2ece333d3f190a2e9b7bcdec4948f555daeb263334c98d239
+size 157680