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": 75833.99938370033,
-      "required_pages_per_second_upper_estimate": 0.006573098867064879
+      "required_token_positions_per_second": 75993.28992288621,
+      "required_pages_per_second_upper_estimate": 0.006586905767283331
     },
     {
       "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": 9479.249922962541,
-      "required_pages_per_second_upper_estimate": 0.00378141452168603
+      "required_token_positions_per_second": 9499.161240360776,
+      "required_pages_per_second_upper_estimate": 0.0037893574438969107
     },
     {
       "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": 18958.499845925082,
-      "required_pages_per_second_upper_estimate": 0.004385715281153322
+      "required_token_positions_per_second": 18998.322480721552,
+      "required_pages_per_second_upper_estimate": 0.00439492754685905
     },
     {
       "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": 28437.749768887625,
-      "required_pages_per_second_upper_estimate": 0.0017317097595831064
+      "required_token_positions_per_second": 28497.483721082324,
+      "required_pages_per_second_upper_estimate": 0.0017353472438719371
     },
     {
       "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": 28437.749768887625,
-      "required_pages_per_second_upper_estimate": 0.0021892166979386807
+      "required_token_positions_per_second": 28497.483721082324,
+      "required_pages_per_second_upper_estimate": 0.0021938151829327913
     },
     {
       "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": 28437.749768887625,
-      "required_pages_per_second_upper_estimate": 0.3531016771035378
+      "required_token_positions_per_second": 28497.483721082324,
+      "required_pages_per_second_upper_estimate": 0.35384337287311823
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0922185412891281,
-    "producer_p95_seconds": 0.19825296900125977,
-    "gpu_update_seconds": 0.9722288234721141,
-    "producer_batches_per_second": 10.843806310758602,
-    "gpu_batches_per_second": 1.0285644447659008,
-    "producer_headroom": 10.542661051468322,
+    "producer_batch_seconds": 0.08068263692972266,
+    "producer_p95_seconds": 0.2145050020008057,
+    "gpu_update_seconds": 0.9701909217881619,
+    "producer_batches_per_second": 12.394240422149739,
+    "gpu_batches_per_second": 1.0307249609766467,
+    "producer_headroom": 12.024779540029552,
     "sustainable": true,
-    "buffer_seconds": 60.278187055271076,
+    "buffer_seconds": 60.15183715086604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244590.10454350527,
-    "upload_bytes_per_second": 5181144.230163514,
+    "generation_bytes_per_second": 257220.49158727934,
+    "upload_bytes_per_second": 5140988.734384648,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.48334266880864,
+    "estimated_full_shards_per_hour": 55.193529708040096,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 9530498,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.9772154700003739,
+      "bytes": 2553165,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 163800, "loss": 0.6356027200818062, "lr": 0.0003, "grad_norm": 0.1450797198632695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23973143738, "tokens_per_second": 123202.7238449831, "data_wait_s": 1.3926000974606723e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163900, "loss": 0.6405609054490924, "lr": 0.0003, "grad_norm": 0.15846700923920673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23987641182, "tokens_per_second": 148192.35858095007, "data_wait_s": 9.921999662765302e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 164000, "loss": 0.6417503990232944, "lr": 0.0003, "grad_norm": 0.13953640851368884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24002151901, "tokens_per_second": 148670.25178825393, "data_wait_s": 1.195699951495044e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 164100, "loss": 0.5901363529264927, "lr": 0.0003, "grad_norm": 0.15520978738963395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24016819212, "tokens_per_second": 150447.37663213024, "data_wait_s": 1.1328998880344443e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c5475eaf4e32d11a7c06db2d51f3a2c84c6d39f6eb30ca7d5bdce7eb1e10cff
-size 191568
+oid sha256:495c361f4125643d89d04d04723d90f51316e4bae96d563f304296642ca4f5a3
+size 193686