Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75475.41268904982,
-      "required_pages_per_second_upper_estimate": 0.006542017481202242
+      "required_token_positions_per_second": 75724.46690303588,
+      "required_pages_per_second_upper_estimate": 0.0065636048692484196
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9434.426586131227,
-      "required_pages_per_second_upper_estimate": 0.0037635338224554123
+      "required_token_positions_per_second": 9465.558362879485,
+      "required_pages_per_second_upper_estimate": 0.003775952753661834
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18868.853172262454,
-      "required_pages_per_second_upper_estimate": 0.004364977100929044
+      "required_token_positions_per_second": 18931.11672575897,
+      "required_pages_per_second_upper_estimate": 0.004379380678229347
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28303.279758393677,
-      "required_pages_per_second_upper_estimate": 0.0017235212414536453
+      "required_token_positions_per_second": 28396.675088638454,
+      "required_pages_per_second_upper_estimate": 0.0017292085270581242
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28303.279758393677,
-      "required_pages_per_second_upper_estimate": 0.002178864824294043
+      "required_token_positions_per_second": 28396.675088638454,
+      "required_pages_per_second_upper_estimate": 0.0021860546553511074
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28303.279758393677,
-      "required_pages_per_second_upper_estimate": 0.35143200961537774
+      "required_token_positions_per_second": 28396.675088638454,
+      "required_pages_per_second_upper_estimate": 0.35259166704295486
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.4853144441171935,
-    "producer_p95_seconds": 0.21884791900083655,
-    "gpu_update_seconds": 0.9768479213721036,
-    "producer_batches_per_second": 2.0605197560502044,
-    "gpu_batches_per_second": 1.0237008014465305,
-    "producer_headroom": 2.012814440643796,
+    "producer_batch_seconds": 0.08333823112495509,
+    "producer_p95_seconds": 0.28399352800079214,
+    "gpu_update_seconds": 0.9736351144526072,
+    "producer_batches_per_second": 11.999294759455923,
+    "gpu_batches_per_second": 1.0270788154166106,
+    "producer_headroom": 11.682934726473437,
     "sustainable": true,
-    "buffer_seconds": 60.56457112507042,
+    "buffer_seconds": 60.36537709606165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250000.7059543864,
+    "generation_bytes_per_second": 278477.4236821034,
     "upload_bytes_per_second": 5805216.83603484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.644331779229105,
+    "estimated_full_shards_per_hour": 59.75477249953581,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 368700, "loss": 0.562000397592783, "lr": 0.0003, "grad_norm": 0.14543126631900252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53967315924, "tokens_per_second": 148058.31468745435, "data_wait_s": 1.321800118603278e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368800, "loss": 0.5869172811508179, "lr": 0.0003, "grad_norm": 0.15332136855723302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53982090871, "tokens_per_second": 150386.07507671177, "data_wait_s": 1.2170999980298802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368900, "loss": 0.5812274729833007, "lr": 0.0003, "grad_norm": 0.1567969206131114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53996816231, "tokens_per_second": 150214.356637518, "data_wait_s": 1.3823999324813485e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 369000, "loss": 0.6362853767350316, "lr": 0.0003, "grad_norm": 0.1764599757176835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54011481448, "tokens_per_second": 149399.63492308176, "data_wait_s": 1.6088999473140575e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c9cf47be5dded19e8eb06b28449902768bc506ace7dc7fb2567c1f80c35bc5a2
-size 296133
+oid sha256:c1167faa1aa4ae19f2695cfa6dd9ade1475894998fdd5d13dace061a59d815ad
+size 298251