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": 74893.80031258588,
-      "required_pages_per_second_upper_estimate": 0.0064916047944934396
+      "required_token_positions_per_second": 74959.82335637519,
+      "required_pages_per_second_upper_estimate": 0.006497327504595222
     },
     {
       "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": 9361.725039073235,
-      "required_pages_per_second_upper_estimate": 0.003734532088348985
+      "required_token_positions_per_second": 9369.977919546898,
+      "required_pages_per_second_upper_estimate": 0.003737824285761488
     },
     {
       "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": 18723.45007814647,
-      "required_pages_per_second_upper_estimate": 0.004331340654112366
+      "required_token_positions_per_second": 18739.955839093796,
+      "required_pages_per_second_upper_estimate": 0.004335158971416059
     },
     {
       "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": 28085.175117219707,
-      "required_pages_per_second_upper_estimate": 0.001710239813112766
+      "required_token_positions_per_second": 28109.9337586407,
+      "required_pages_per_second_upper_estimate": 0.001711747484476752
     },
     {
       "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": 28085.175117219707,
-      "required_pages_per_second_upper_estimate": 0.0021620745252641824
+      "required_token_positions_per_second": 28109.9337586407,
+      "required_pages_per_second_upper_estimate": 0.002163980514017076
     },
     {
       "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": 28085.175117219707,
-      "required_pages_per_second_upper_estimate": 0.3487238799212748
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28109.9337586407,
+      "required_pages_per_second_upper_estimate": 0.349031299385881
     }
   ],
   "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": 0.0793058158593567,
-    "producer_p95_seconds": 0.28030279200174846,
-    "gpu_update_seconds": 0.9844339543764616,
-    "producer_batches_per_second": 12.609415705065437,
-    "gpu_batches_per_second": 1.015812178718884,
-    "producer_headroom": 12.413136964914228,
+    "producer_batch_seconds": 0.12030383693746671,
+    "producer_p95_seconds": 0.31131690699839965,
+    "gpu_update_seconds": 0.983566885549892,
+    "producer_batches_per_second": 8.312286835205386,
+    "gpu_batches_per_second": 1.0167076735619465,
+    "producer_headroom": 8.17569007430033,
     "sustainable": true,
-    "buffer_seconds": 60.05047121696416,
+    "buffer_seconds": 60.98114690409331,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287132.20739136037,
-    "upload_bytes_per_second": 5306007.97869946,
+    "generation_bytes_per_second": 317058.9679943056,
+    "upload_bytes_per_second": 5246174.938218343,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.61188761048897,
+    "estimated_full_shards_per_hour": 68.03347377654912,
     "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": 0.9567362410016358,
-      "bytes": 6277518,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2580463369995414,
       "bytes": 7425150,
@@ -562,6 +556,12 @@
       "bytes": 6182999,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8964433900000586,
+      "bytes": 1180194,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 338900, "loss": 0.5484571140259504, "lr": 0.0003, "grad_norm": 0.1442884114183441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49605719347, "tokens_per_second": 147974.9637834689, "data_wait_s": 1.793300179997459e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339000, "loss": 0.6315003428608179, "lr": 0.0003, "grad_norm": 0.1682675100232813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49620516109, "tokens_per_second": 149760.62995231096, "data_wait_s": 1.692300065769814e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339100, "loss": 0.6156422272324562, "lr": 0.0003, "grad_norm": 0.16153957412485814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49635297006, "tokens_per_second": 150227.11843808705, "data_wait_s": 1.4793000445934013e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 339200, "loss": 0.5894141495227814, "lr": 0.0003, "grad_norm": 0.15781046016238667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49649972896, "tokens_per_second": 148500.65319454408, "data_wait_s": 1.6009998944355175e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:713538d31eb8de6ecdbbde345b612bdbff22adb2e1f388403e2aacebedb51984
-size 513735
+oid sha256:dc0b6e083c38c2b43a13f198db77b23c3608e67b7f200917efced61e6702984a
+size 515853