Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75936.78517276923,
-      "required_pages_per_second_upper_estimate": 0.005253482995869071
+      "required_token_positions_per_second": 75991.51670516684,
+      "required_pages_per_second_upper_estimate": 0.0052572694502749885
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9492.098146596154,
-      "required_pages_per_second_upper_estimate": 0.00397414327643044
+      "required_token_positions_per_second": 9498.939588145855,
+      "required_pages_per_second_upper_estimate": 0.0039770076451417545
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18984.196293192308,
-      "required_pages_per_second_upper_estimate": 0.003953907440948659
+      "required_token_positions_per_second": 18997.87917629171,
+      "required_pages_per_second_upper_estimate": 0.003956757224656372
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28476.29443978846,
-      "required_pages_per_second_upper_estimate": 0.0017071103708813278
+      "required_token_positions_per_second": 28496.818764437565,
+      "required_pages_per_second_upper_estimate": 0.0017083407728052124
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28476.29443978846,
-      "required_pages_per_second_upper_estimate": 0.002213906638711561
+      "required_token_positions_per_second": 28496.818764437565,
+      "required_pages_per_second_upper_estimate": 0.0022155023146760654
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28476.29443978846,
-      "required_pages_per_second_upper_estimate": 0.3652633296108113
+      "required_token_positions_per_second": 28496.818764437565,
+      "required_pages_per_second_upper_estimate": 0.36552659361010714
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09621346602358472,
-    "producer_p95_seconds": 0.27603234599882853,
-    "gpu_update_seconds": 0.9709128432584568,
-    "producer_batches_per_second": 10.393555510773004,
-    "gpu_batches_per_second": 1.029958566253923,
-    "producer_headroom": 10.09123653252922,
+    "producer_batch_seconds": 0.09569582548422773,
+    "producer_p95_seconds": 0.27176082499863696,
+    "gpu_update_seconds": 0.9702135606274465,
+    "producer_batches_per_second": 10.449776622333612,
+    "gpu_batches_per_second": 1.0307009101720763,
+    "producer_headroom": 10.138514984515744,
     "sustainable": true,
-    "buffer_seconds": 60.19659628202432,
+    "buffer_seconds": 60.15324075890168,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364325.4061968371,
-    "upload_bytes_per_second": 4834978.016490366,
+    "generation_bytes_per_second": 349582.5161070806,
+    "upload_bytes_per_second": 4743179.573413656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.17575110844454,
+    "estimated_full_shards_per_hour": 75.01227009209933,
     "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": 1.2419597409971175,
-      "bytes": 13297063,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.2846820170016144,
       "bytes": 6246650,
@@ -562,6 +556,12 @@
       "bytes": 5484455,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.2610901230000309,
+      "bytes": 5822650,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 715000, "loss": 0.49898889660835266, "lr": 8e-05, "grad_norm": 0.23847586120249772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104656097809, "tokens_per_second": 152420.1536742034, "data_wait_s": 1.3430999388219789e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715100, "loss": 0.5293847918510437, "lr": 8e-05, "grad_norm": 0.2190585582182101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104670736248, "tokens_per_second": 134843.6941909461, "data_wait_s": 6.782000127714127e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 715200, "loss": 0.5196549892425537, "lr": 8e-05, "grad_norm": 0.21338338280575464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104685388793, "tokens_per_second": 152555.09060464893, "data_wait_s": 1.3106997357681394e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715300, "loss": 0.5784443020820618, "lr": 8e-05, "grad_norm": 0.2182984198762408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104700026515, "tokens_per_second": 152363.27701295802, "data_wait_s": 1.5354002243839204e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37a1597e0ddd71b01a723ac3ac0a0c1588d16194d79a7f367c1d5f9d3debd11c
-size 644499
+oid sha256:afe86d9a08e4a0ff1acd32fbbe604ef39032efcd3aa865c0411ec7d93fdf7a65
+size 646617