Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75716.55208952597,
-      "required_pages_per_second_upper_estimate": 0.006562918833273292
+      "required_token_positions_per_second": 75751.09065223065,
+      "required_pages_per_second_upper_estimate": 0.006565912548351346
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9464.569011190746,
-      "required_pages_per_second_upper_estimate": 0.0037755580864810697
+      "required_token_positions_per_second": 9468.88633152883,
+      "required_pages_per_second_upper_estimate": 0.003777280330113623
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18929.138022381492,
-      "required_pages_per_second_upper_estimate": 0.004378922940026958
+      "required_token_positions_per_second": 18937.77266305766,
+      "required_pages_per_second_upper_estimate": 0.004380920412182904
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28393.707033572235,
-      "required_pages_per_second_upper_estimate": 0.0017290277880767728
+      "required_token_positions_per_second": 28406.65899458649,
+      "required_pages_per_second_upper_estimate": 0.0017298164945488534
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28393.707033572235,
-      "required_pages_per_second_upper_estimate": 0.0021858261662560073
+      "required_token_positions_per_second": 28406.65899458649,
+      "required_pages_per_second_upper_estimate": 0.0021868232440682076
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28393.707033572235,
-      "required_pages_per_second_upper_estimate": 0.3525548137324737
+      "required_token_positions_per_second": 28406.65899458649,
+      "required_pages_per_second_upper_estimate": 0.3527156337408457
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10614924216457666,
-    "producer_p95_seconds": 0.35374181999941356,
-    "gpu_update_seconds": 0.9737368906183851,
-    "producer_batches_per_second": 9.420698439368723,
-    "gpu_batches_per_second": 1.0269714638878846,
-    "producer_headroom": 9.173281605804373,
+    "producer_batch_seconds": 0.10508160640648612,
+    "producer_p95_seconds": 0.35450192500138655,
+    "gpu_update_seconds": 0.9732929171737137,
+    "producer_batches_per_second": 9.51641333052818,
+    "gpu_batches_per_second": 1.027439923125958,
+    "producer_headroom": 9.26225769150059,
     "sustainable": true,
-    "buffer_seconds": 60.37168721833987,
+    "buffer_seconds": 60.344160864770245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243617.83686030228,
-    "upload_bytes_per_second": 5674282.793405771,
+    "generation_bytes_per_second": 253692.02681536527,
+    "upload_bytes_per_second": 5607397.059808457,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.27471665722658,
+    "estimated_full_shards_per_hour": 54.436403306443395,
     "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": 1.2118571440005326,
-      "bytes": 6204328,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1613783780012454,
       "bytes": 6337641,
@@ -562,6 +556,12 @@
       "bytes": 6218344,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1409971860048245,
+      "bytes": 830736,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -277,3 +277,4 @@
 {"step": 307900, "loss": 0.5711268819868565, "lr": 0.0003, "grad_norm": 0.16283465958012144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45068149169, "tokens_per_second": 118495.73769390934, "data_wait_s": 9.321003744844347e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 308000, "loss": 0.5821310188621283, "lr": 0.0003, "grad_norm": 0.1793600506390255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45082883292, "tokens_per_second": 150587.10168041612, "data_wait_s": 1.3660996046382934e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 308100, "loss": 0.5738916657865047, "lr": 0.0003, "grad_norm": 0.15274501562727477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45097640050, "tokens_per_second": 151146.1361488563, "data_wait_s": 1.0962998203467578e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 308200, "loss": 0.6042536422610283, "lr": 0.0003, "grad_norm": 0.15310585430395826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45112381171, "tokens_per_second": 151511.93431939805, "data_wait_s": 9.064999176189303e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2ff3b01ba9e2678e3604827f138eb08caa343c49fb9fdb87fb98a5cbd044524b
-size 595002
+oid sha256:8f6a4d7ed3d67a8035866cdd4fc385969c1405232a0d7d94d4bc8666fd27df82
+size 597120