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": 76395.04643687642,
-      "required_pages_per_second_upper_estimate": 0.005285186573432635
+      "required_token_positions_per_second": 76740.89208576432,
+      "required_pages_per_second_upper_estimate": 0.0053091129778951645
     },
     {
       "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": 9549.380804609553,
-      "required_pages_per_second_upper_estimate": 0.003998126329142726
+      "required_token_positions_per_second": 9592.61151072054,
+      "required_pages_per_second_upper_estimate": 0.004016226123031597
     },
     {
       "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": 19098.761609219106,
-      "required_pages_per_second_upper_estimate": 0.003977768374986484
+      "required_token_positions_per_second": 19185.22302144108,
+      "required_pages_per_second_upper_estimate": 0.003995776007011549
     },
     {
       "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": 28648.14241382866,
-      "required_pages_per_second_upper_estimate": 0.0017174123945283746
+      "required_token_positions_per_second": 28777.83453216162,
+      "required_pages_per_second_upper_estimate": 0.0017251872389940372
     },
     {
       "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": 28648.14241382866,
-      "required_pages_per_second_upper_estimate": 0.0022272670628138325
+      "required_token_positions_per_second": 28777.83453216162,
+      "required_pages_per_second_upper_estimate": 0.0022373500545588795
     },
     {
       "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": 28648.14241382866,
-      "required_pages_per_second_upper_estimate": 0.36746761090582036
+      "required_token_positions_per_second": 28777.83453216162,
+      "required_pages_per_second_upper_estimate": 0.3691311621472483
     }
   ],
   "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": 0.1325808826015873,
-    "producer_p95_seconds": 0.3024487070069881,
-    "gpu_update_seconds": 0.9650887516760639,
-    "producer_batches_per_second": 7.5425655673529794,
-    "gpu_batches_per_second": 1.036174132444613,
-    "producer_headroom": 7.27924518783155,
+    "producer_batch_seconds": 0.1181306096877961,
+    "producer_p95_seconds": 0.31607507199805696,
+    "gpu_update_seconds": 0.960739418009408,
+    "producer_batches_per_second": 8.465206457859402,
+    "gpu_batches_per_second": 1.040864964270892,
+    "producer_headroom": 8.132857525653325,
     "sustainable": true,
-    "buffer_seconds": 60.800591355592026,
+    "buffer_seconds": 60.5265833345927,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393897.51257035317,
-    "upload_bytes_per_second": 5767579.018418025,
+    "generation_bytes_per_second": 520330.82799151685,
+    "upload_bytes_per_second": 5799792.100889264,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.5212367327971,
+    "estimated_full_shards_per_hour": 111.65088300522928,
     "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": 1.295129406000342,
-      "bytes": 6202154,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2203903480003646,
       "bytes": 12408575,
@@ -562,6 +556,12 @@
       "bytes": 875423,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.1449231579972547,
+      "bytes": 7698354,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -357,3 +357,4 @@
 {"step": 720700, "loss": 0.5245054364204407, "lr": 8e-05, "grad_norm": 0.2286034159329082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105490324667, "tokens_per_second": 150841.09056894577, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720800, "loss": 0.5011938214302063, "lr": 8e-05, "grad_norm": 0.2403712256520014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105504891856, "tokens_per_second": 151785.59926973566, "data_wait_s": 9.578994649928063e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 720900, "loss": 0.5309969186782837, "lr": 8e-05, "grad_norm": 0.21036737659257107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105519523252, "tokens_per_second": 152467.02468842303, "data_wait_s": 1.1761003406718373e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721000, "loss": 0.5678845643997192, "lr": 8e-05, "grad_norm": 0.21371042788309177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105534247906, "tokens_per_second": 152453.4650900134, "data_wait_s": 1.1078001989517361e-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:c8ea83ee80293969752d730385fbc0b4eeddeaede02c61341762b0af71539cb2
-size 766008
+oid sha256:33e62a05dae09cd142cecd9f84bfcca50b09c9513b33fe27dd33b0634de118fc
+size 768126