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": 75086.7229248902,
-      "required_pages_per_second_upper_estimate": 0.006508326837570087
+      "required_token_positions_per_second": 75186.55382029519,
+      "required_pages_per_second_upper_estimate": 0.006516979926564701
     },
     {
       "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": 9385.840365611275,
-      "required_pages_per_second_upper_estimate": 0.0037441520526612713
+      "required_token_positions_per_second": 9398.319227536898,
+      "required_pages_per_second_upper_estimate": 0.0037491300572590146
     },
     {
       "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": 18771.68073122255,
-      "required_pages_per_second_upper_estimate": 0.0043424979668710145
+      "required_token_positions_per_second": 18796.638455073797,
+      "required_pages_per_second_upper_estimate": 0.004348271497043062
     },
     {
       "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": 28157.52109683382,
-      "required_pages_per_second_upper_estimate": 0.0017146453037012452
+      "required_token_positions_per_second": 28194.957682610693,
+      "required_pages_per_second_upper_estimate": 0.001716924995360471
     },
     {
       "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": 28157.52109683382,
-      "required_pages_per_second_upper_estimate": 0.002167643919041366
+      "required_token_positions_per_second": 28194.957682610693,
+      "required_pages_per_second_upper_estimate": 0.002170525891045572
     },
     {
       "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": 28157.52109683382,
-      "required_pages_per_second_upper_estimate": 0.34962217486166386
+      "required_token_positions_per_second": 28194.957682610693,
+      "required_pages_per_second_upper_estimate": 0.35008701196481984
     }
   ],
   "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": 0.5280077561408234,
-    "producer_p95_seconds": 0.34268352299841354,
-    "gpu_update_seconds": 0.9819046181273707,
-    "producer_batches_per_second": 1.8939115730210845,
-    "gpu_batches_per_second": 1.0184288591158066,
-    "producer_headroom": 1.8596405198742758,
+    "producer_batch_seconds": 0.11327465061771136,
+    "producer_p95_seconds": 0.35754993600130547,
+    "gpu_update_seconds": 0.9806008688231502,
+    "producer_batches_per_second": 8.828100502158092,
+    "gpu_batches_per_second": 1.0197829022935,
+    "producer_headroom": 8.656843022474312,
     "sustainable": true,
-    "buffer_seconds": 60.87808632389699,
+    "buffer_seconds": 60.797253867035316,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395772.92164319876,
-    "upload_bytes_per_second": 4040033.856714506,
+    "generation_bytes_per_second": 395611.21062026295,
+    "upload_bytes_per_second": 4153866.550660145,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.92365586254093,
+    "estimated_full_shards_per_hour": 84.88895644145886,
     "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": 1.124874531000387,
-      "bytes": 2649444,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9738625030004187,
       "bytes": 6071559,
@@ -562,6 +556,12 @@
       "bytes": 878239,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.8046116949990392,
+      "bytes": 17147881,
+      "pages": 44,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 109500, "loss": 0.5849812235683203, "lr": 0.0003, "grad_norm": 0.13039552184598213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16023487806, "tokens_per_second": 146737.0379731941, "data_wait_s": 1.657399843679741e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109600, "loss": 0.5687562748789787, "lr": 0.0003, "grad_norm": 0.1292022099549797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16038041293, "tokens_per_second": 137484.17929221195, "data_wait_s": 1.445900124963373e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109700, "loss": 0.6325525287538767, "lr": 0.0003, "grad_norm": 0.13897102358023364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16052565456, "tokens_per_second": 147192.1196538167, "data_wait_s": 1.6413003322668374e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 109800, "loss": 0.5832493174821138, "lr": 0.0003, "grad_norm": 0.16027330536996562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16067239572, "tokens_per_second": 148674.31861393442, "data_wait_s": 1.4754004951100796e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f9582d6993599257316969841878e17a4f85e573069a00345997d3969c993ff
-size 526443
+oid sha256:08f9141b18a0cbf9d50e1c1616e6463c6b1acd836ba3748b30c29ac9948af332
+size 528561