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": 77017.34182130285,
-      "required_pages_per_second_upper_estimate": 0.0053282384120514805
+      "required_token_positions_per_second": 76858.30253805296,
+      "required_pages_per_second_upper_estimate": 0.005317235705414275
     },
     {
       "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": 9627.167727662856,
-      "required_pages_per_second_upper_estimate": 0.004030694089449477
+      "required_token_positions_per_second": 9607.28781725662,
+      "required_pages_per_second_upper_estimate": 0.004022370786102123
     },
     {
       "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": 19254.335455325712,
-      "required_pages_per_second_upper_estimate": 0.00401017030437227
+      "required_token_positions_per_second": 19214.57563451324,
+      "required_pages_per_second_upper_estimate": 0.004001889382234018
     },
     {
       "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": 28881.50318298857,
-      "required_pages_per_second_upper_estimate": 0.0017314020163182483
+      "required_token_positions_per_second": 28821.86345176986,
+      "required_pages_per_second_upper_estimate": 0.0017278267054962834
     },
     {
       "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": 28881.50318298857,
-      "required_pages_per_second_upper_estimate": 0.00224540983617047
+      "required_token_positions_per_second": 28821.86345176986,
+      "required_pages_per_second_upper_estimate": 0.0022407731093956693
     },
     {
       "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": 28881.50318298857,
-      "required_pages_per_second_upper_estimate": 0.3704609122893315
+      "required_token_positions_per_second": 28821.86345176986,
+      "required_pages_per_second_upper_estimate": 0.3696959178534121
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.048069187710929384,
-    "producer_p95_seconds": 0.16215918899979442,
-    "gpu_update_seconds": 0.9572908939270477,
-    "producer_batches_per_second": 20.80334716728804,
-    "gpu_batches_per_second": 1.044614553782862,
-    "producer_headroom": 19.914854806447885,
+    "producer_batch_seconds": 0.06703939030467865,
+    "producer_p95_seconds": 0.16387904899966088,
+    "gpu_update_seconds": 0.9592717711075764,
+    "producer_batches_per_second": 14.916603439488776,
+    "gpu_batches_per_second": 1.042457445448852,
+    "producer_headroom": 14.309076600307764,
     "sustainable": true,
-    "buffer_seconds": 60.30932631740401,
+    "buffer_seconds": 60.43412157977732,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346976.92967743875,
-    "upload_bytes_per_second": 6005378.700800214,
+    "generation_bytes_per_second": 346195.87268839625,
+    "upload_bytes_per_second": 5985267.879030746,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.45317189924594,
+    "estimated_full_shards_per_hour": 74.28557525147357,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 856165,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.0899097739993522,
+      "bytes": 6114737,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 487500, "loss": 0.5860530138015747, "lr": 0.0003, "grad_norm": 0.16962373406060718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71353551408, "tokens_per_second": 149781.97212477008, "data_wait_s": 1.2200000128359534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 487600, "loss": 0.5476698875427246, "lr": 0.0003, "grad_norm": 0.13413675514355675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71368311775, "tokens_per_second": 153787.58064907705, "data_wait_s": 1.158499981102068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 487700, "loss": 0.6029510498046875, "lr": 0.0003, "grad_norm": 0.16517729298767686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71383011697, "tokens_per_second": 152984.05714421393, "data_wait_s": 1.2878999768872745e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 487800, "loss": 0.5893129110336304, "lr": 0.0003, "grad_norm": 0.15771968225877409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71397697664, "tokens_per_second": 152379.4101436609, "data_wait_s": 1.1095999980170745e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9993967641d6a1a407db0ba5846605de52521ffe79c006984021aa419aea6493
-size 57351
+oid sha256:d35ca5f3a2bc890f50979d6f330cd035cd1dc2820cd1aab8fd99263745b0cebc
+size 59469