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": 74887.08250471727,
-      "required_pages_per_second_upper_estimate": 0.006491022511933516
+      "required_token_positions_per_second": 75186.27613883483,
+      "required_pages_per_second_upper_estimate": 0.006516955857839494
     },
     {
       "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": 9360.885313089659,
-      "required_pages_per_second_upper_estimate": 0.0037341971090991137
+      "required_token_positions_per_second": 9398.284517354354,
+      "required_pages_per_second_upper_estimate": 0.003749116210848234
     },
     {
       "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": 18721.770626179317,
-      "required_pages_per_second_upper_estimate": 0.00433095214245711
+      "required_token_positions_per_second": 18796.569034708707,
+      "required_pages_per_second_upper_estimate": 0.0043482554378633555
     },
     {
       "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": 28082.655939268974,
-      "required_pages_per_second_upper_estimate": 0.0017100864083926717
+      "required_token_positions_per_second": 28194.85355206306,
+      "required_pages_per_second_upper_estimate": 0.001716918654356451
     },
     {
       "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": 28082.655939268974,
-      "required_pages_per_second_upper_estimate": 0.002161880591972004
+      "required_token_positions_per_second": 28194.85355206306,
+      "required_pages_per_second_upper_estimate": 0.0021705178747877633
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28082.655939268974,
-      "required_pages_per_second_upper_estimate": 0.34869260016227294
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28194.85355206306,
+      "required_pages_per_second_upper_estimate": 0.35008571901192076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15211595453109794,
-    "producer_p95_seconds": 0.3268552970002929,
-    "gpu_update_seconds": 0.984522263841107,
-    "producer_batches_per_second": 6.573932386530594,
-    "gpu_batches_per_second": 1.0157210626182356,
-    "producer_headroom": 6.472182795525471,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.055858094307524,
+    "buffer_seconds": 60.79747840628777,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331871.3488988658,
-    "upload_bytes_per_second": 4775567.4506229935,
+    "generation_bytes_per_second": 311342.4758280202,
+    "upload_bytes_per_second": 4838145.632111306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.21186590408783,
+    "estimated_full_shards_per_hour": 66.80684763079122,
     "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": 1.288112906000606,
-      "bytes": 5599143,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.0466477340014535,
       "bytes": 7416772,
@@ -562,6 +556,12 @@
       "bytes": 5697842,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1368190850007522,
+      "bytes": 10318857,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -306,3 +306,4 @@
 {"step": 267700, "loss": 0.5945382248610258, "lr": 0.0003, "grad_norm": 0.15191960258312273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39183308033, "tokens_per_second": 149902.31310225226, "data_wait_s": 1.5765999705763534e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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:d8394e24928a540c12252063e1935556ee415b592386d2e20a16aa055e2b2cd7
-size 657207
+oid sha256:61d97978a7eb80ef9ab5835ef604e1695dbafb1bd244468f8dde401a0e253266
+size 659325