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": 75595.77596003689,
-      "required_pages_per_second_upper_estimate": 0.00655245026447348
+      "required_token_positions_per_second": 75683.75558037784,
+      "required_pages_per_second_upper_estimate": 0.006560076114982325
     },
     {
       "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": 9449.471995004611,
-      "required_pages_per_second_upper_estimate": 0.0037695356610039136
+      "required_token_positions_per_second": 9460.46944754723,
+      "required_pages_per_second_upper_estimate": 0.0037739227092497326
     },
     {
       "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": 18898.943990009222,
-      "required_pages_per_second_upper_estimate": 0.004371938082034454
+      "required_token_positions_per_second": 18920.93889509446,
+      "required_pages_per_second_upper_estimate": 0.004377026216228814
     },
     {
       "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": 28348.415985013835,
-      "required_pages_per_second_upper_estimate": 0.0017262698008433336
+      "required_token_positions_per_second": 28381.408342641687,
+      "required_pages_per_second_upper_estimate": 0.001728278862325347
     },
     {
       "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": 28348.415985013835,
-      "required_pages_per_second_upper_estimate": 0.002182339535964335
+      "required_token_positions_per_second": 28381.408342641687,
+      "required_pages_per_second_upper_estimate": 0.0021848793789832183
     },
     {
       "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": 28348.415985013835,
-      "required_pages_per_second_upper_estimate": 0.3519924504887671
+      "required_token_positions_per_second": 28381.408342641687,
+      "required_pages_per_second_upper_estimate": 0.35240210515218706
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10761398057027804,
-    "producer_p95_seconds": 0.35367215199948987,
-    "gpu_update_seconds": 0.9752925882919136,
-    "producer_batches_per_second": 9.29247291755873,
-    "gpu_batches_per_second": 1.025333332791299,
-    "producer_headroom": 9.062879963398363,
+    "producer_batch_seconds": 0.08291008243759279,
+    "producer_p95_seconds": 0.3194867050005996,
+    "gpu_update_seconds": 0.9741588460379615,
+    "producer_batches_per_second": 12.06125926545435,
+    "gpu_batches_per_second": 1.026526632763371,
+    "producer_headroom": 11.74958240779968,
     "sustainable": true,
-    "buffer_seconds": 60.46814047409864,
+    "buffer_seconds": 60.397848454353614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287941.1146042781,
-    "upload_bytes_per_second": 3929122.153925739,
+    "generation_bytes_per_second": 310868.0856826792,
+    "upload_bytes_per_second": 3883003.3917949684,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.78546026798494,
+    "estimated_full_shards_per_hour": 66.70505454883845,
     "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": 1.0121167409997724,
-      "bytes": 6826316,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.0728631890015095,
       "bytes": 9404752,
@@ -562,6 +556,12 @@
       "bytes": 7611348,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8344998959983059,
+      "bytes": 1093697,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 142800, "loss": 0.5808074194937944, "lr": 0.0003, "grad_norm": 0.15253148407926956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20898723309, "tokens_per_second": 147789.7326634042, "data_wait_s": 9.75099828792736e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 142900, "loss": 0.6225517578423023, "lr": 0.0003, "grad_norm": 0.13541117666423091, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20913218827, "tokens_per_second": 148693.3120301591, "data_wait_s": 1.123200127040036e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143000, "loss": 0.5215986203402281, "lr": 0.0003, "grad_norm": 0.1486694121606959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20927832816, "tokens_per_second": 150014.96200869096, "data_wait_s": 1.2985998182557523e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 143100, "loss": 0.5752749182283878, "lr": 0.0003, "grad_norm": 0.1271331963668286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20942532801, "tokens_per_second": 150436.25070818825, "data_wait_s": 1.4188000932335854e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:687c47264dd60c7afffb7362a7a2a84dd3f3d658933063bcf83e0eb7b48c9ee1
-size 597120
+oid sha256:994c17b6b2c331def0320027c0bbf942caff6d63af7046b2edfa850ce6a63bfe
+size 599238