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": 76963.25328375488,
-      "required_pages_per_second_upper_estimate": 0.005324496441521214
+      "required_token_positions_per_second": 76927.34033656325,
+      "required_pages_per_second_upper_estimate": 0.005322011900504979
     },
     {
       "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": 9620.40666046936,
-      "required_pages_per_second_upper_estimate": 0.0040278633718026985
+      "required_token_positions_per_second": 9615.917542070407,
+      "required_pages_per_second_upper_estimate": 0.004025983871672509
     },
     {
       "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": 19240.81332093872,
-      "required_pages_per_second_upper_estimate": 0.0040073540003821235
+      "required_token_positions_per_second": 19231.835084140814,
+      "required_pages_per_second_upper_estimate": 0.004005484070429146
     },
     {
       "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": 28861.21998140808,
-      "required_pages_per_second_upper_estimate": 0.0017301860693541536
+      "required_token_positions_per_second": 28847.75262621122,
+      "required_pages_per_second_upper_estimate": 0.0017293787219735625
     },
     {
       "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": 28861.21998140808,
-      "required_pages_per_second_upper_estimate": 0.0022438329064639623
+      "required_token_positions_per_second": 28847.75262621122,
+      "required_pages_per_second_upper_estimate": 0.0022427858788340413
     },
     {
       "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": 28861.21998140808,
-      "required_pages_per_second_upper_estimate": 0.370200741157862
+      "required_token_positions_per_second": 28847.75262621122,
+      "required_pages_per_second_upper_estimate": 0.3700279963855161
     }
   ],
   "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": 0.10813953361730455,
-    "producer_p95_seconds": 0.3106010260016774,
-    "gpu_update_seconds": 0.957963662582884,
-    "producer_batches_per_second": 9.247311936251771,
-    "gpu_batches_per_second": 1.0438809310405122,
-    "producer_headroom": 8.858588811498167,
+    "producer_batch_seconds": 0.7866119535858616,
+    "producer_p95_seconds": 0.3272384250012692,
+    "gpu_update_seconds": 0.9584108806756365,
+    "producer_batches_per_second": 1.2712748585136349,
+    "gpu_batches_per_second": 1.0433938305197923,
+    "producer_headroom": 1.2184036567288479,
     "sustainable": true,
-    "buffer_seconds": 60.35171074272169,
+    "buffer_seconds": 60.3798854825651,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338338.05841589055,
-    "upload_bytes_per_second": 4890686.467958432,
+    "generation_bytes_per_second": 296091.28241508733,
+    "upload_bytes_per_second": 4955363.003361818,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.59947122914826,
+    "estimated_full_shards_per_hour": 63.53429655398812,
     "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": 1.0188206910024746,
-      "bytes": 812811,
-      "pages": 23,
-      "shards": 1
-    },
     {
       "seconds": 1.1825939109985484,
       "bytes": 10557366,
@@ -562,6 +556,12 @@
       "bytes": 6081250,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.8620396420010366,
+      "bytes": 10625025,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -372,3 +372,4 @@
 {"step": 757200, "loss": 0.5424105525016785, "lr": 5e-05, "grad_norm": 0.2516967332565844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110832886816, "tokens_per_second": 153187.1995101724, "data_wait_s": 1.166600122814998e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757300, "loss": 0.5943609476089478, "lr": 5e-05, "grad_norm": 0.25545669909743174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110847362637, "tokens_per_second": 150128.62361775842, "data_wait_s": 1.0565003321971744e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757400, "loss": 0.5036542415618896, "lr": 5e-05, "grad_norm": 0.24222789077215542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110861845054, "tokens_per_second": 133779.16123272188, "data_wait_s": 1.686399627942592e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 757500, "loss": 0.5535759925842285, "lr": 5e-05, "grad_norm": 0.2478978984704597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110876418893, "tokens_per_second": 150270.35519568075, "data_wait_s": 1.3442004274111241e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:787a0c8eee59cc9339600929c4be3ad07128cc5269beabdddc45a31e8b842d45
-size 797778
+oid sha256:ef28018079c7ea796ab904e24b29905de39550cc7ed74b1f96c69776fcdb7814
+size 799896