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": 75186.27613883483,
-      "required_pages_per_second_upper_estimate": 0.006516955857839494
+      "required_token_positions_per_second": 75067.74118622203,
+      "required_pages_per_second_upper_estimate": 0.006506681548571107
     },
     {
       "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": 9398.284517354354,
-      "required_pages_per_second_upper_estimate": 0.003749116210848234
+      "required_token_positions_per_second": 9383.467648277754,
+      "required_pages_per_second_upper_estimate": 0.0037432055402416443
     },
     {
       "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": 18796.569034708707,
-      "required_pages_per_second_upper_estimate": 0.0043482554378633555
+      "required_token_positions_per_second": 18766.935296555508,
+      "required_pages_per_second_upper_estimate": 0.004341400194077591
     },
     {
       "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": 28194.85355206306,
-      "required_pages_per_second_upper_estimate": 0.001716918654356451
+      "required_token_positions_per_second": 28150.402944833262,
+      "required_pages_per_second_upper_estimate": 0.0017142118455904685
     },
     {
       "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": 28194.85355206306,
-      "required_pages_per_second_upper_estimate": 0.0021705178747877633
+      "required_token_positions_per_second": 28150.402944833262,
+      "required_pages_per_second_upper_estimate": 0.002167095943996057
     },
     {
       "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": 28194.85355206306,
-      "required_pages_per_second_upper_estimate": 0.35008571901192076
+      "required_token_positions_per_second": 28150.402944833262,
+      "required_pages_per_second_upper_estimate": 0.34953379123673917
     }
   ],
   "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": 0.4386131674140188,
-    "producer_p95_seconds": 0.3431156429978728,
-    "gpu_update_seconds": 0.9806044904239963,
-    "producer_batches_per_second": 2.2799133138109213,
-    "gpu_batches_per_second": 1.0197791359976511,
-    "producer_headroom": 2.2356932333004433,
+    "producer_batch_seconds": 0.08530850346889451,
+    "producer_p95_seconds": 0.305827556996519,
+    "gpu_update_seconds": 0.9821529039631217,
+    "producer_batches_per_second": 11.722160855448877,
+    "gpu_batches_per_second": 1.018171402807916,
+    "producer_headroom": 11.512954324901946,
     "sustainable": true,
-    "buffer_seconds": 60.79747840628777,
+    "buffer_seconds": 60.893480045713545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311342.4758280202,
-    "upload_bytes_per_second": 4838145.632111306,
+    "generation_bytes_per_second": 311341.15884218534,
+    "upload_bytes_per_second": 4850527.454510985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.80684763079122,
+    "estimated_full_shards_per_hour": 66.80656503628893,
     "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": 1.0466477340014535,
-      "bytes": 7416772,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.263268196998979,
       "bytes": 1135440,
@@ -562,6 +556,12 @@
       "bytes": 10318857,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0222451929985255,
+      "bytes": 8375215,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -307,3 +307,4 @@
 {"step": 267800, "loss": 0.5631391536444426, "lr": 0.0003, "grad_norm": 0.14669065333577141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39197877179, "tokens_per_second": 148099.78003376513, "data_wait_s": 1.5671001165173948e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 267900, "loss": 0.6203124411404133, "lr": 0.0003, "grad_norm": 0.1908258995673714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39212530776, "tokens_per_second": 149008.2329960209, "data_wait_s": 1.1150998034281656e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268000, "loss": 0.5497099291533232, "lr": 0.0003, "grad_norm": 0.14825696052239232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39227199752, "tokens_per_second": 149621.39817186145, "data_wait_s": 1.5390000044135377e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 268100, "loss": 0.5864941235631704, "lr": 0.0003, "grad_norm": 0.14459926591310038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39241925666, "tokens_per_second": 149693.1115663491, "data_wait_s": 1.1463998816907406e-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 sha256:61d97978a7eb80ef9ab5835ef604e1695dbafb1bd244468f8dde401a0e253266
-size 659325
+oid sha256:afb1130950a4ec6b61d07fa0cb38d74e60fb2c7ae321249a50f4c57b73a8e5d5
+size 661443