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": 75694.85170177964,
-      "required_pages_per_second_upper_estimate": 0.0052367454769830174
+      "required_token_positions_per_second": 75996.84948531997,
+      "required_pages_per_second_upper_estimate": 0.005257638384379737
     },
     {
       "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": 9461.856462722455,
-      "required_pages_per_second_upper_estimate": 0.003961481714910693
+      "required_token_positions_per_second": 9499.606185664996,
+      "required_pages_per_second_upper_estimate": 0.00397728673560288
     },
     {
       "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": 18923.71292544491,
-      "required_pages_per_second_upper_estimate": 0.003941310350500546
+      "required_token_positions_per_second": 18999.21237132999,
+      "required_pages_per_second_upper_estimate": 0.003957034894024128
     },
     {
       "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": 28385.569388167365,
-      "required_pages_per_second_upper_estimate": 0.0017016715425657758
+      "required_token_positions_per_second": 28498.81855699499,
+      "required_pages_per_second_upper_estimate": 0.0017084606573155236
     },
     {
       "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": 28385.569388167365,
-      "required_pages_per_second_upper_estimate": 0.002206853165005349
+      "required_token_positions_per_second": 28498.81855699499,
+      "required_pages_per_second_upper_estimate": 0.0022156577897512473
     },
     {
       "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": 28385.569388167365,
-      "required_pages_per_second_upper_estimate": 0.36409960606158676
+      "required_token_positions_per_second": 28498.81855699499,
+      "required_pages_per_second_upper_estimate": 0.36555224480182386
     }
   ],
   "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": 0.08327861103896339,
-    "producer_p95_seconds": 0.2283959929991397,
-    "gpu_update_seconds": 0.9740160439242476,
-    "producer_batches_per_second": 12.00788518833644,
-    "gpu_batches_per_second": 1.026677133541933,
-    "producer_headroom": 11.695872827040027,
+    "producer_batch_seconds": 0.3809065089063495,
+    "producer_p95_seconds": 0.22878464799941867,
+    "gpu_update_seconds": 0.9701454797049418,
+    "producer_batches_per_second": 2.6253161251331156,
+    "gpu_batches_per_second": 1.0307732406320524,
+    "producer_headroom": 2.5469385715943855,
     "sustainable": true,
-    "buffer_seconds": 60.38899472330335,
+    "buffer_seconds": 60.14901974170639,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381375.99750558473,
-    "upload_bytes_per_second": 5810614.77646332,
+    "generation_bytes_per_second": 341620.3187043019,
+    "upload_bytes_per_second": 5889029.72166122,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.8344110858503,
+    "estimated_full_shards_per_hour": 73.30376787993234,
     "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": 1.1130377059998864,
-      "bytes": 986874,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.09502404000159,
       "bytes": 13537871,
@@ -562,6 +556,12 @@
       "bytes": 2381673,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0147103390008851,
+      "bytes": 6222923,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -209,3 +209,4 @@
 {"step": 705900, "loss": 0.5515061020851135, "lr": 8e-05, "grad_norm": 0.21426312437840853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103323815108, "tokens_per_second": 151287.48383465843, "data_wait_s": 1.9654002244351432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706000, "loss": 0.5447523593902588, "lr": 8e-05, "grad_norm": 0.2117613878313574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103338378624, "tokens_per_second": 151162.66131677548, "data_wait_s": 1.6653997590765357e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706100, "loss": 0.5687307715415955, "lr": 8e-05, "grad_norm": 0.21028804868749565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103353132130, "tokens_per_second": 153307.28542239554, "data_wait_s": 1.4781999198021367e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 706200, "loss": 0.52701735496521, "lr": 8e-05, "grad_norm": 0.20526991568566821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103367686119, "tokens_per_second": 152454.54314131514, "data_wait_s": 1.6400001186411828e-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 sha256:9a6f2ea98af516721764f4656614d3964633b4f316ab52fce9fbebe6379df791
-size 450195
+oid sha256:2f00a812c3b1738e81f99cec199407c0080a9fa182fc234d78d9b02714a2d98a
+size 452313