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": 76633.31485835434,
-      "required_pages_per_second_upper_estimate": 0.005301670535689384
+      "required_token_positions_per_second": 76984.2191212589,
+      "required_pages_per_second_upper_estimate": 0.005325946906807187
     },
     {
       "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": 9579.164357294292,
-      "required_pages_per_second_upper_estimate": 0.004010596080700523
+      "required_token_positions_per_second": 9623.027390157362,
+      "required_pages_per_second_upper_estimate": 0.004028960616596006
     },
     {
       "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": 19158.328714588584,
-      "required_pages_per_second_upper_estimate": 0.00399017463214474
+      "required_token_positions_per_second": 19246.054780314724,
+      "required_pages_per_second_upper_estimate": 0.004008445658143566
     },
     {
       "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": 28737.49307188288,
-      "required_pages_per_second_upper_estimate": 0.0017227688300481697
+      "required_token_positions_per_second": 28869.082170472084,
+      "required_pages_per_second_upper_estimate": 0.001730657395583673
     },
     {
       "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": 28737.49307188288,
-      "required_pages_per_second_upper_estimate": 0.002234213683465538
+      "required_token_positions_per_second": 28869.082170472084,
+      "required_pages_per_second_upper_estimate": 0.0022444441570815734
     },
     {
       "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": 28737.49307188288,
-      "required_pages_per_second_upper_estimate": 0.36861370521007786
+      "required_token_positions_per_second": 28869.082170472084,
+      "required_pages_per_second_upper_estimate": 0.37030158887741416
     }
   ],
   "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": 0.0909844579688297,
-    "producer_p95_seconds": 0.27212792099817307,
-    "gpu_update_seconds": 0.9620880962317185,
-    "producer_batches_per_second": 10.990888139846799,
-    "gpu_batches_per_second": 1.0394058547411342,
-    "producer_headroom": 10.574202646360982,
+    "producer_batch_seconds": 0.18400648166411315,
+    "producer_p95_seconds": 0.26011092200133135,
+    "gpu_update_seconds": 0.9577027713156384,
+    "producer_batches_per_second": 5.434591167420981,
+    "gpu_batches_per_second": 1.044165298411172,
+    "producer_headroom": 5.204723022006564,
     "sustainable": true,
-    "buffer_seconds": 60.61155006259827,
+    "buffer_seconds": 60.33527459288522,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328801.4256086898,
-    "upload_bytes_per_second": 3595181.968850426,
+    "generation_bytes_per_second": 330756.23118513095,
+    "upload_bytes_per_second": 3612608.3228906672,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.55313183017273,
+    "estimated_full_shards_per_hour": 70.97258760133215,
     "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": 1.1801181450009608,
-      "bytes": 7664644,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0472080479994474,
       "bytes": 5473896,
@@ -562,6 +556,12 @@
       "bytes": 6771836,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.460754736999661,
+      "bytes": 10771768,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 863900, "loss": 0.5542579889297485, "lr": 3e-05, "grad_norm": 0.3171998236379697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126450890686, "tokens_per_second": 151451.18440000675, "data_wait_s": 9.624000085750595e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 864000, "loss": 0.5007447004318237, "lr": 3e-05, "grad_norm": 0.3006990500875119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126465423623, "tokens_per_second": 151092.7277189139, "data_wait_s": 9.35500065679662e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 864100, "loss": 0.5419777631759644, "lr": 3e-05, "grad_norm": 0.31882175322727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126479984323, "tokens_per_second": 150390.26237038162, "data_wait_s": 1.6076999600045383e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 864200, "loss": 0.5264800786972046, "lr": 3e-05, "grad_norm": 0.31084709395988436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126494659108, "tokens_per_second": 152552.89609263383, "data_wait_s": 1.3656001101480797e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e45386ade555b5c8d586a4bfb1d32788d7da27a2bcc54f7d90ec403b48565eae
-size 460785
+oid sha256:bffc199edc06fd63bcb49f504078661ade2dc48b886ea24ace9d3fd1be9170ec
+size 462903