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": 76860.48684863307,
-      "required_pages_per_second_upper_estimate": 0.005317386821088506
+      "required_token_positions_per_second": 76668.92268928129,
+      "required_pages_per_second_upper_estimate": 0.005304133968054461
     },
     {
       "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": 9607.560856079133,
-      "required_pages_per_second_upper_estimate": 0.00402248510175541
+      "required_token_positions_per_second": 9583.615336160161,
+      "required_pages_per_second_upper_estimate": 0.004012459612604653
     },
     {
       "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": 19215.121712158267,
-      "required_pages_per_second_upper_estimate": 0.004002003115806441
+      "required_token_positions_per_second": 19167.230672320322,
+      "required_pages_per_second_upper_estimate": 0.0039920286751798486
     },
     {
       "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": 28822.6825682374,
-      "required_pages_per_second_upper_estimate": 0.001727875810277794
+      "required_token_positions_per_second": 28750.846008480483,
+      "required_pages_per_second_upper_estimate": 0.0017235693182084426
     },
     {
       "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": 28822.6825682374,
-      "required_pages_per_second_upper_estimate": 0.002240836792098108
+      "required_token_positions_per_second": 28750.846008480483,
+      "required_pages_per_second_upper_estimate": 0.0022352518155526408
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28822.6825682374,
-      "required_pages_per_second_upper_estimate": 0.36970642459995895
+      "required_token_positions_per_second": 28750.846008480483,
+      "required_pages_per_second_upper_estimate": 0.3687849823434856
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10967683465624134,
-    "producer_p95_seconds": 0.23130529400077648,
-    "gpu_update_seconds": 0.9592445094082984,
-    "producer_batches_per_second": 9.117695665947023,
-    "gpu_batches_per_second": 1.0424870720571977,
-    "producer_headroom": 8.746099506015522,
+    "producer_batch_seconds": 0.07884335089846672,
+    "producer_p95_seconds": 0.2173753309998574,
+    "gpu_update_seconds": 0.9616412675941716,
+    "producer_batches_per_second": 12.683377718024502,
+    "gpu_batches_per_second": 1.0398888168576563,
+    "producer_headroom": 12.196859426136754,
     "sustainable": true,
-    "buffer_seconds": 60.4324040927228,
+    "buffer_seconds": 60.583399858432806,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454130.9820771153,
-    "upload_bytes_per_second": 4008657.877863186,
+    "generation_bytes_per_second": 454630.07360466046,
+    "upload_bytes_per_second": 4011505.656124086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.44593712554067,
+    "estimated_full_shards_per_hour": 97.55303054909572,
     "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": 1.6070335669992346,
-      "bytes": 5667674,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.649759572000221,
       "bytes": 7616382,
@@ -562,6 +556,12 @@
       "bytes": 7118698,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.6447518009990745,
+      "bytes": 6129393,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 812900, "loss": 0.5041095614433289, "lr": 3e-05, "grad_norm": 0.2833444949896702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118985207065, "tokens_per_second": 150032.91286971106, "data_wait_s": 2.9550999897764996e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813000, "loss": 0.5881887078285217, "lr": 3e-05, "grad_norm": 0.32019736391769926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118999865026, "tokens_per_second": 151582.16613952265, "data_wait_s": 1.4104998626862653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813100, "loss": 0.5329916477203369, "lr": 3e-05, "grad_norm": 0.291605878109602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119014428690, "tokens_per_second": 150627.43206019083, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 813200, "loss": 0.5572542548179626, "lr": 3e-05, "grad_norm": 0.3302031908063002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119029057222, "tokens_per_second": 151758.50868091194, "data_wait_s": 1.0109000868396834e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4a4e3641241bcc1e23ba8320c757acd03be7d3753432917192257a198047954
-size 279189
+oid sha256:cec6cd1b3d68a5e5efca08b030fc9a7b0f52de191afd7632e3496b2cb62f6a10
+size 281307