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": 76220.52180059816,
-      "required_pages_per_second_upper_estimate": 0.005273112554141956
+      "required_token_positions_per_second": 76682.26579525918,
+      "required_pages_per_second_upper_estimate": 0.005305057075086283
     },
     {
       "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": 9527.56522507477,
-      "required_pages_per_second_upper_estimate": 0.003988992601552617
+      "required_token_positions_per_second": 9585.283224407398,
+      "required_pages_per_second_upper_estimate": 0.004013157922584305
     },
     {
       "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": 19055.13045014954,
-      "required_pages_per_second_upper_estimate": 0.003968681155183335
+      "required_token_positions_per_second": 19170.566448814796,
+      "required_pages_per_second_upper_estimate": 0.0039927234294533055
     },
     {
       "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": 28582.695675224306,
-      "required_pages_per_second_upper_estimate": 0.0017134889624804272
+      "required_token_positions_per_second": 28755.849673222194,
+      "required_pages_per_second_upper_estimate": 0.0017238692802695544
     },
     {
       "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": 28582.695675224306,
-      "required_pages_per_second_upper_estimate": 0.002222178866757124
+      "required_token_positions_per_second": 28755.849673222194,
+      "required_pages_per_second_upper_estimate": 0.0022356408284775137
     },
     {
       "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": 28582.695675224306,
-      "required_pages_per_second_upper_estimate": 0.3666281304142367
+      "required_token_positions_per_second": 28755.849673222194,
+      "required_pages_per_second_upper_estimate": 0.3688491639822757
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07521880509384005,
+    "producer_batch_seconds": 0.2727295860001675,
     "producer_p95_seconds": 0.2716047489993798,
-    "gpu_update_seconds": 0.9672985471403767,
-    "producer_batches_per_second": 13.294547802938894,
-    "gpu_batches_per_second": 1.0338069905680087,
-    "producer_headroom": 12.859796774671079,
+    "gpu_update_seconds": 0.9614739371010887,
+    "producer_batches_per_second": 3.6666355662615415,
+    "gpu_batches_per_second": 1.0400697943150388,
+    "producer_headroom": 3.5253745338083644,
     "sustainable": true,
-    "buffer_seconds": 60.93980846984373,
+    "buffer_seconds": 60.57285803736859,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421740.73238193186,
-    "upload_bytes_per_second": 5632818.257623442,
+    "generation_bytes_per_second": 417531.3331951958,
+    "upload_bytes_per_second": 5633362.077796813,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.49574354737727,
+    "estimated_full_shards_per_hour": 89.59250447170167,
     "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": 1.609685786999762,
-      "bytes": 5649551,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 0.9700031609972939,
       "bytes": 2540670,
@@ -562,6 +556,12 @@
       "bytes": 7685246,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8166221829997085,
+      "bytes": 1223594,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -229,3 +229,4 @@
 {"step": 507900, "loss": 0.5805848240852356, "lr": 0.0003, "grad_norm": 0.19505596456534124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74339433629, "tokens_per_second": 150216.0210410353, "data_wait_s": 1.0264000593451783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 508000, "loss": 0.5835515856742859, "lr": 0.0003, "grad_norm": 0.18415195227724593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74354071869, "tokens_per_second": 152597.08112120087, "data_wait_s": 7.861002814024687e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 508100, "loss": 0.6056517362594604, "lr": 0.0003, "grad_norm": 0.16544024449387015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74368753894, "tokens_per_second": 153186.17762033496, "data_wait_s": 8.586001058574766e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 508200, "loss": 0.5989865064620972, "lr": 0.0003, "grad_norm": 0.1722002195965079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74383504709, "tokens_per_second": 153661.78541964802, "data_wait_s": 8.799001079751179e-06, "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 sha256:9c8cbd0b97087e75b38e97a1bd0166144ded307ca6607f03a13d3cf2236a240f
-size 492555
+oid sha256:e3bca6a94148274b317177a71ce54556e2c089636ebe8f51a1844e7435a7a4fd
+size 494673