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": 75862.18750229837,
-      "required_pages_per_second_upper_estimate": 0.006575542141742825
+      "required_token_positions_per_second": 75845.13063578129,
+      "required_pages_per_second_upper_estimate": 0.006574063695783357
     },
     {
       "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": 9482.773437787297,
-      "required_pages_per_second_upper_estimate": 0.003782820104430867
+      "required_token_positions_per_second": 9480.64132947266,
+      "required_pages_per_second_upper_estimate": 0.0037819695745462986
     },
     {
       "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": 18965.546875574593,
-      "required_pages_per_second_upper_estimate": 0.00438734548743925
+      "required_token_positions_per_second": 18961.28265894532,
+      "required_pages_per_second_upper_estimate": 0.004386359035969718
     },
     {
       "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": 28448.320313361888,
-      "required_pages_per_second_upper_estimate": 0.0017323534502822282
+      "required_token_positions_per_second": 28441.92398841798,
+      "required_pages_per_second_upper_estimate": 0.0017319639476520677
     },
     {
       "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": 28448.320313361888,
-      "required_pages_per_second_upper_estimate": 0.002190030447713449
+      "required_token_positions_per_second": 28441.92398841798,
+      "required_pages_per_second_upper_estimate": 0.0021895380409131067
     },
     {
       "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": 28448.320313361888,
-      "required_pages_per_second_upper_estimate": 0.3532329278885716
+      "required_token_positions_per_second": 28441.92398841798,
+      "required_pages_per_second_upper_estimate": 0.3531535069398907
     }
   ],
   "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": 0.11627025199993568,
-    "producer_p95_seconds": 0.26189117499961867,
-    "gpu_update_seconds": 0.9718675723365647,
-    "producer_batches_per_second": 8.600652211543785,
-    "gpu_batches_per_second": 1.0289467705932396,
-    "producer_headroom": 8.358694985344165,
+    "producer_batch_seconds": 0.09667869972639664,
+    "producer_p95_seconds": 0.25321765299668186,
+    "gpu_update_seconds": 0.9720861363408017,
+    "producer_batches_per_second": 10.343540023087064,
+    "gpu_batches_per_second": 1.0287154220347938,
+    "producer_headroom": 10.054811857129149,
     "sustainable": true,
-    "buffer_seconds": 60.255789484867016,
+    "buffer_seconds": 60.2693404531297,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385701.97946807713,
-    "upload_bytes_per_second": 3946120.9874398075,
+    "generation_bytes_per_second": 386312.180201163,
+    "upload_bytes_per_second": 3956709.8652422405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.76266611129509,
+    "estimated_full_shards_per_hour": 82.89360098386923,
     "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": 0.9350508080005966,
-      "bytes": 6046815,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3128581109995139,
       "bytes": 2374553,
@@ -562,6 +556,12 @@
       "bytes": 6005453,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.00243121099993,
+      "bytes": 7452907,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 293100, "loss": 0.5831968560814857, "lr": 0.0003, "grad_norm": 0.15030196289962997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42901558737, "tokens_per_second": 150688.91296613295, "data_wait_s": 1.6109999705804512e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293200, "loss": 0.5806980840861797, "lr": 0.0003, "grad_norm": 0.14956070229021243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42916220789, "tokens_per_second": 150175.45408458274, "data_wait_s": 1.2565000361064449e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293300, "loss": 0.5915417168289423, "lr": 0.0003, "grad_norm": 0.1528488919679106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42930985270, "tokens_per_second": 152309.3695231858, "data_wait_s": 1.55180023284629e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 293400, "loss": 0.6367223151028156, "lr": 0.0003, "grad_norm": 0.1769232731724887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42945773257, "tokens_per_second": 150833.97254747324, "data_wait_s": 7.929000275908038e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e49e61a87963eeb25284f052f104faff13ed59cad604d9e727ff5bbddbd4c123
-size 279189
+oid sha256:551bb76e3a043532f3dcaef0fc0685bbbbb510940276457aaa6b0d46ea8bae36
+size 281307