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": 75305.4116003562,
-      "required_pages_per_second_upper_estimate": 0.006527282217698099
+      "required_token_positions_per_second": 75139.87038762507,
+      "required_pages_per_second_upper_estimate": 0.006512933524938933
     },
     {
       "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": 9413.176450044524,
-      "required_pages_per_second_upper_estimate": 0.003755056825452579
+      "required_token_positions_per_second": 9392.483798453133,
+      "required_pages_per_second_upper_estimate": 0.003746802217350061
     },
     {
       "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": 18826.35290008905,
-      "required_pages_per_second_upper_estimate": 0.00435514541094097
+      "required_token_positions_per_second": 18784.967596906266,
+      "required_pages_per_second_upper_estimate": 0.004345571649406093
     },
     {
       "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": 28239.529350133573,
-      "required_pages_per_second_upper_estimate": 0.0017196391760631476
+      "required_token_positions_per_second": 28177.4513953594,
+      "required_pages_per_second_upper_estimate": 0.001715858954315258
     },
     {
       "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": 28239.529350133573,
-      "required_pages_per_second_upper_estimate": 0.0021739571413937554
+      "required_token_positions_per_second": 28177.4513953594,
+      "required_pages_per_second_upper_estimate": 0.0021691782085924622
     },
     {
       "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": 28239.529350133573,
-      "required_pages_per_second_upper_estimate": 0.35064044290367874
+      "required_token_positions_per_second": 28177.4513953594,
+      "required_pages_per_second_upper_estimate": 0.3498696424669332
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.06384072853906275,
-    "producer_p95_seconds": 0.1551805370004331,
-    "gpu_update_seconds": 0.9790531441654223,
-    "producer_batches_per_second": 15.663981644384302,
-    "gpu_batches_per_second": 1.021395014110734,
-    "producer_headroom": 15.335870479083914,
+    "producer_batch_seconds": 0.4284042593281434,
+    "producer_p95_seconds": 0.15817792999996527,
+    "gpu_update_seconds": 0.9812101034997581,
+    "producer_batches_per_second": 2.3342438321418117,
+    "gpu_batches_per_second": 1.0191497177140987,
+    "producer_headroom": 2.290383632129539,
     "sustainable": true,
-    "buffer_seconds": 60.70129493825618,
+    "buffer_seconds": 60.835026416985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418550.5217334221,
-    "upload_bytes_per_second": 4449513.535273651,
+    "generation_bytes_per_second": 341719.08924540004,
+    "upload_bytes_per_second": 4732867.436690909,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.81119860651015,
+    "estimated_full_shards_per_hour": 73.32496173879147,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2577832,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.8447169340001892,
+      "bytes": 6262278,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 358300, "loss": 0.5646369121968746, "lr": 0.0003, "grad_norm": 0.15815820866515076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52445227242, "tokens_per_second": 148204.9657593968, "data_wait_s": 9.793000117497286e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 358400, "loss": 0.5718606263399124, "lr": 0.0003, "grad_norm": 0.14460316342390767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52459947917, "tokens_per_second": 150141.10795959993, "data_wait_s": 1.5287000223906944e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358500, "loss": 0.5853067152202129, "lr": 0.0003, "grad_norm": 0.1982320209235525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52474704097, "tokens_per_second": 150531.51776591752, "data_wait_s": 1.1948000064876396e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 358600, "loss": 0.5974908228963614, "lr": 0.0003, "grad_norm": 0.16560434333579174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52489362573, "tokens_per_second": 149132.2637041259, "data_wait_s": 6.9150000854278915e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65374e8d65bcfae66ec47f557f6b22ebc4cba959714991c5d8fda2965469eb96
-size 74295
+oid sha256:13784ccf3994584bcffdc255f68db94cf86e9d8946c50d6805dba531070c7135
+size 76413