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": 76642.27563507736,
-      "required_pages_per_second_upper_estimate": 0.005302290463014949
+      "required_token_positions_per_second": 76675.80365820727,
+      "required_pages_per_second_upper_estimate": 0.005304610009450808
     },
     {
       "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": 9580.28445438467,
-      "required_pages_per_second_upper_estimate": 0.004011065041961976
+      "required_token_positions_per_second": 9584.475457275908,
+      "required_pages_per_second_upper_estimate": 0.004012819727354446
     },
     {
       "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": 19160.56890876934,
-      "required_pages_per_second_upper_estimate": 0.003990641205514698
+      "required_token_positions_per_second": 19168.950914551817,
+      "required_pages_per_second_upper_estimate": 0.003992386956270831
     },
     {
       "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": 28740.853363154012,
-      "required_pages_per_second_upper_estimate": 0.0017229702743790066
+      "required_token_positions_per_second": 28753.426371827725,
+      "required_pages_per_second_upper_estimate": 0.001723724007051123
     },
     {
       "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": 28740.853363154012,
-      "required_pages_per_second_upper_estimate": 0.0022344749313315095
+      "required_token_positions_per_second": 28753.426371827725,
+      "required_pages_per_second_upper_estimate": 0.00223545242745307
     },
     {
       "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": 28740.853363154012,
-      "required_pages_per_second_upper_estimate": 0.36865680741850426
+      "required_token_positions_per_second": 28753.426371827725,
+      "required_pages_per_second_upper_estimate": 0.3688180804739257
     }
   ],
   "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": 0.09345906353149758,
-    "producer_p95_seconds": 0.269405070997891,
-    "gpu_update_seconds": 0.9619756118809243,
-    "producer_batches_per_second": 10.699871817814438,
-    "gpu_batches_per_second": 1.0395273930538922,
-    "producer_headroom": 10.293015738989501,
+    "producer_batch_seconds": 0.08131561201554405,
+    "producer_p95_seconds": 0.26547220800057403,
+    "gpu_update_seconds": 0.9615549688745684,
+    "producer_batches_per_second": 12.297761465643806,
+    "gpu_batches_per_second": 1.0399821459717782,
+    "producer_headroom": 11.824973643323995,
     "sustainable": true,
-    "buffer_seconds": 60.60446354849823,
+    "buffer_seconds": 60.57796303909781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 261856.68537681297,
-    "upload_bytes_per_second": 5259035.961411815,
+    "generation_bytes_per_second": 283466.0060889987,
+    "upload_bytes_per_second": 5266049.741023806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.188348970206185,
+    "estimated_full_shards_per_hour": 60.82520615580054,
     "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": 1.4115151900005003,
-      "bytes": 7731251,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1964002810018428,
       "bytes": 11692696,
@@ -562,6 +556,12 @@
       "bytes": 1827152,
       "pages": 11,
       "shards": 2
+    },
+    {
+      "seconds": 0.9918526799992833,
+      "bytes": 6095255,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -300,3 +300,4 @@
 {"step": 750000, "loss": 0.550243616104126, "lr": 5e-05, "grad_norm": 0.24450628622732434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109778926010, "tokens_per_second": 151990.02936366422, "data_wait_s": 1.3694996596314013e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750100, "loss": 0.5139143466949463, "lr": 5e-05, "grad_norm": 0.23780783135987257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109793600219, "tokens_per_second": 139187.46416654414, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750200, "loss": 0.5065025091171265, "lr": 5e-05, "grad_norm": 0.24197594067939596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109808240721, "tokens_per_second": 151238.24581371917, "data_wait_s": 1.4543002180289477e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 750300, "loss": 0.5430476665496826, "lr": 5e-05, "grad_norm": 0.24048518397270588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109822774191, "tokens_per_second": 149618.8203046912, "data_wait_s": 1.6566002159379423e-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:c3d7ae5860f6f5eaf918b80e60f2231c1cedd7a90adb15547c905701b69f09af
-size 644499
+oid sha256:dd6e4d2fa2ed82c4f3cf65f5378b2ad75c1d6db6d2757cdd0e7fb9954056e027
+size 646617