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": 76744.0656468477,
-      "required_pages_per_second_upper_estimate": 0.005309332532214588
+      "required_token_positions_per_second": 76751.17991915937,
+      "required_pages_per_second_upper_estimate": 0.0053098247140804964
     },
     {
       "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": 9593.008205855962,
-      "required_pages_per_second_upper_estimate": 0.0040163922110008615
+      "required_token_positions_per_second": 9593.897489894922,
+      "required_pages_per_second_upper_estimate": 0.004016764535657612
     },
     {
       "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": 19186.016411711924,
-      "required_pages_per_second_upper_estimate": 0.003995941249281857
+      "required_token_positions_per_second": 19187.794979789844,
+      "required_pages_per_second_upper_estimate": 0.003996311678108493
     },
     {
       "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": 28779.024617567888,
-      "required_pages_per_second_upper_estimate": 0.0017252585827969877
+      "required_token_positions_per_second": 28781.692469684764,
+      "required_pages_per_second_upper_estimate": 0.00172541851645782
     },
     {
       "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": 28779.024617567888,
-      "required_pages_per_second_upper_estimate": 0.002237442578464584
+      "required_token_positions_per_second": 28781.692469684764,
+      "required_pages_per_second_upper_estimate": 0.00223764999223203
     },
     {
       "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": 28779.024617567888,
-      "required_pages_per_second_upper_estimate": 0.36914642728502567
+      "required_token_positions_per_second": 28781.692469684764,
+      "required_pages_per_second_upper_estimate": 0.36918064762746455
     }
   ],
   "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": 0.5998053112109858,
-    "producer_p95_seconds": 0.27605460099948687,
-    "gpu_update_seconds": 0.9606996890062263,
-    "producer_batches_per_second": 1.6672076443955377,
-    "gpu_batches_per_second": 1.0409080084479125,
-    "producer_headroom": 1.601685865479596,
+    "producer_batch_seconds": 0.09606207005469969,
+    "producer_p95_seconds": 0.27300766600092174,
+    "gpu_update_seconds": 0.9606106391804838,
+    "producer_batches_per_second": 10.409935986498935,
+    "gpu_batches_per_second": 1.0410045019417233,
+    "producer_headroom": 9.999895261818663,
     "sustainable": true,
-    "buffer_seconds": 60.524080407392255,
+    "buffer_seconds": 60.51847026837048,
     "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": 234892.18194810566,
-    "upload_bytes_per_second": 4626268.275509967,
+    "generation_bytes_per_second": 235141.59924877153,
+    "upload_bytes_per_second": 4647383.918342561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.402394235919736,
+    "estimated_full_shards_per_hour": 50.45591338250503,
     "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": 1.4291834129980998,
-      "bytes": 4870018,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.6493652240023948,
       "bytes": 6440305,
@@ -562,6 +556,12 @@
       "bytes": 6359348,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.232538223001029,
+      "bytes": 5896962,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 914700, "loss": 0.6132705807685852, "lr": 3e-05, "grad_norm": 0.36856665212879103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133885605938, "tokens_per_second": 154593.4702946775, "data_wait_s": 1.4293000276666135e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914800, "loss": 0.5447489023208618, "lr": 3e-05, "grad_norm": 0.34278712036988007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133900311472, "tokens_per_second": 126826.12141213288, "data_wait_s": 1.4359000488184392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914900, "loss": 0.5532621741294861, "lr": 3e-05, "grad_norm": 0.33788388176720446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133914855868, "tokens_per_second": 151628.29680967875, "data_wait_s": 9.654999303165823e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 915000, "loss": 0.5263814926147461, "lr": 3e-05, "grad_norm": 0.31579094226674975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133929383665, "tokens_per_second": 136591.20693712513, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58751d3a47878855b8ef6f6f721bd467d8ae1879488be7da8f1cee75d9ea68a8
-size 637362
+oid sha256:f0e03cd9fa4e39dc43dc7824e44c0c3b17658647b17dac0a09fc1661a33cf5d6
+size 639480