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": 77196.27955109101,
-      "required_pages_per_second_upper_estimate": 0.0053406177393909605
+      "required_token_positions_per_second": 76886.94193362926,
+      "required_pages_per_second_upper_estimate": 0.0053192170452527965
     },
     {
       "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": 9649.534943886376,
-      "required_pages_per_second_upper_estimate": 0.004040058775801676
+      "required_token_positions_per_second": 9610.867741703658,
+      "required_pages_per_second_upper_estimate": 0.004023869625710774
     },
     {
       "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": 19299.069887772752,
-      "required_pages_per_second_upper_estimate": 0.004019487306924672
+      "required_token_positions_per_second": 19221.735483407316,
+      "required_pages_per_second_upper_estimate": 0.004003380589940741
     },
     {
       "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": 28948.604831659126,
-      "required_pages_per_second_upper_estimate": 0.0017354246576977632
+      "required_token_positions_per_second": 28832.603225110975,
+      "required_pages_per_second_upper_estimate": 0.001728470538509399
     },
     {
       "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": 28948.604831659126,
-      "required_pages_per_second_upper_estimate": 0.0022506266942056456
+      "required_token_positions_per_second": 28832.603225110975,
+      "required_pages_per_second_upper_estimate": 0.002241608079533671
     },
     {
       "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": 28948.604831659126,
-      "required_pages_per_second_upper_estimate": 0.3713216201903404
+      "required_token_positions_per_second": 28832.603225110975,
+      "required_pages_per_second_upper_estimate": 0.36983367613436174
     }
   ],
   "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": 0.08863720699201849,
-    "producer_p95_seconds": 0.24648081300256308,
-    "gpu_update_seconds": 0.9550719338903427,
-    "producer_batches_per_second": 11.28194393681704,
-    "gpu_batches_per_second": 1.0470415520710041,
-    "producer_headroom": 10.775068013778277,
+    "producer_batch_seconds": 0.10754288338290507,
+    "producer_p95_seconds": 0.24255180299951462,
+    "gpu_update_seconds": 0.9589144547281365,
+    "producer_batches_per_second": 9.29861622214008,
+    "gpu_batches_per_second": 1.0428458921119421,
+    "producer_headroom": 8.91657750437966,
     "sustainable": true,
-    "buffer_seconds": 60.16953183509159,
+    "buffer_seconds": 60.4116106478726,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343376.9829683435,
-    "upload_bytes_per_second": 5118760.807672537,
+    "generation_bytes_per_second": 382329.7204019299,
+    "upload_bytes_per_second": 5018808.802502223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.6807071379445,
+    "estimated_full_shards_per_hour": 82.03905781787323,
     "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": 1.3904252129996166,
-      "bytes": 12223668,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.4052468339996267,
       "bytes": 11659745,
@@ -562,6 +556,12 @@
       "bytes": 10300175,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.2003908199985744,
+      "bytes": 2637875,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -238,3 +238,4 @@
 {"step": 783800, "loss": 0.5824326276779175, "lr": 3e-05, "grad_norm": 0.2746221799751516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114726173008, "tokens_per_second": 153272.93106106776, "data_wait_s": 1.0961997759295627e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783900, "loss": 0.5601587891578674, "lr": 3e-05, "grad_norm": 0.2715992571080736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114740873998, "tokens_per_second": 153443.8424656105, "data_wait_s": 8.216000423999503e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 784000, "loss": 0.5442437529563904, "lr": 3e-05, "grad_norm": 0.26176771027942847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114755616200, "tokens_per_second": 152848.74576709841, "data_wait_s": 1.0533000022405759e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 784100, "loss": 0.5269461870193481, "lr": 3e-05, "grad_norm": 0.2546602007876645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114770183930, "tokens_per_second": 151698.9817132107, "data_wait_s": 1.3642002159031108e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:045831b3606d947f68daf4d5190d11cde0ae39373c5ae26cf05610a15c5219e2
-size 511617
+oid sha256:e09ae044e7aeafb811790047740f8f7557726054882b693894f37ae0a59e17ba
+size 513735