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": 75476.53556903466,
-      "required_pages_per_second_upper_estimate": 0.006542114809593416
+      "required_token_positions_per_second": 75520.36435578644,
+      "required_pages_per_second_upper_estimate": 0.006545913777746015
     },
     {
       "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": 9434.566946129333,
-      "required_pages_per_second_upper_estimate": 0.0037635898141572255
+      "required_token_positions_per_second": 9440.045544473305,
+      "required_pages_per_second_upper_estimate": 0.003765775308948981
     },
     {
       "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": 18869.133892258666,
-      "required_pages_per_second_upper_estimate": 0.004365042040559664
+      "required_token_positions_per_second": 18880.09108894661,
+      "required_pages_per_second_upper_estimate": 0.0043675767951733445
     },
     {
       "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": 28303.700838388,
-      "required_pages_per_second_upper_estimate": 0.0017235468830160635
+      "required_token_positions_per_second": 28320.136633419916,
+      "required_pages_per_second_upper_estimate": 0.0017245477367015551
     },
     {
       "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": 28303.700838388,
-      "required_pages_per_second_upper_estimate": 0.00217889724019763
+      "required_token_positions_per_second": 28320.136633419916,
+      "required_pages_per_second_upper_estimate": 0.0021801625131963797
     },
     {
       "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": 28303.700838388,
-      "required_pages_per_second_upper_estimate": 0.3514372380196431
+      "required_token_positions_per_second": 28320.136633419916,
+      "required_pages_per_second_upper_estimate": 0.35164131558687206
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.24834316794536448,
-    "producer_p95_seconds": 0.20484636100081843,
-    "gpu_update_seconds": 0.9768333886041263,
-    "producer_batches_per_second": 4.026686170887536,
-    "gpu_batches_per_second": 1.0237160314810474,
-    "producer_headroom": 3.933401497153446,
+    "producer_batch_seconds": 0.059814317031325004,
+    "producer_p95_seconds": 0.18985062799947627,
+    "gpu_update_seconds": 0.9762664763196537,
+    "producer_batches_per_second": 16.71840538572556,
+    "gpu_batches_per_second": 1.0243104974471902,
+    "producer_headroom": 16.321618715605812,
     "sustainable": true,
-    "buffer_seconds": 60.563670093455826,
+    "buffer_seconds": 60.52852153181853,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279931.79001930583,
-    "upload_bytes_per_second": 3689120.873922546,
+    "generation_bytes_per_second": 314721.44035960344,
+    "upload_bytes_per_second": 3697719.0698916214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.06684565958387,
+    "estimated_full_shards_per_hour": 67.53189476100042,
     "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": 1.0558128229999966,
-      "bytes": 6261983,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3157395409998571,
       "bytes": 12201352,
@@ -562,6 +556,12 @@
       "bytes": 1081552,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.197562700999697,
+      "bytes": 7759288,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -94,3 +94,4 @@
 {"step": 246500, "loss": 0.6035121232271194, "lr": 0.0003, "grad_norm": 0.13245181578642634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36080377744, "tokens_per_second": 148907.00192626932, "data_wait_s": 1.5286001143977046e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246600, "loss": 0.5950597636401653, "lr": 0.0003, "grad_norm": 0.1482707455529183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36095092040, "tokens_per_second": 150548.3810541842, "data_wait_s": 1.485900065745227e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246700, "loss": 0.5947669669985771, "lr": 0.0003, "grad_norm": 0.15534384714170624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36109688356, "tokens_per_second": 149075.8801741165, "data_wait_s": 1.2619000699487515e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246800, "loss": 0.6446951851248741, "lr": 0.0003, "grad_norm": 0.14357716821001443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36124434723, "tokens_per_second": 149976.7950049897, "data_wait_s": 1.3602000763057731e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9ee6399dd610df92978433e14d8207de7ba953bfc86c76e310edd0b067b00c2
-size 205059
+oid sha256:1d39209e6e8fa07fcb44e3b4da4777b69505633d86a560dc2c7e6954dcfe4e00
+size 207177