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": 75669.12137117634,
-      "required_pages_per_second_upper_estimate": 0.006558807658818803
+      "required_token_positions_per_second": 75738.33364701176,
+      "required_pages_per_second_upper_estimate": 0.006564806803471308
     },
     {
       "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": 9458.640171397043,
-      "required_pages_per_second_upper_estimate": 0.0037731929836433073
+      "required_token_positions_per_second": 9467.29170587647,
+      "required_pages_per_second_upper_estimate": 0.0037766442100991187
     },
     {
       "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": 18917.280342794085,
-      "required_pages_per_second_upper_estimate": 0.0043761798745424445
+      "required_token_positions_per_second": 18934.58341175294,
+      "required_pages_per_second_upper_estimate": 0.004380182635022462
     },
     {
       "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": 28375.92051419113,
-      "required_pages_per_second_upper_estimate": 0.0017279446823651188
+      "required_token_positions_per_second": 28401.875117629414,
+      "required_pages_per_second_upper_estimate": 0.00172952518154123
     },
     {
       "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": 28375.92051419113,
-      "required_pages_per_second_upper_estimate": 0.002184456910757813
+      "required_token_positions_per_second": 28401.875117629414,
+      "required_pages_per_second_upper_estimate": 0.0021864549679774326
     },
     {
       "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": 28375.92051419113,
-      "required_pages_per_second_upper_estimate": 0.35233396468941147
+      "required_token_positions_per_second": 28401.875117629414,
+      "required_pages_per_second_upper_estimate": 0.3526562339996451
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07606619890634647,
-    "producer_p95_seconds": 0.194992065000406,
-    "gpu_update_seconds": 0.9743472457985253,
-    "producer_batches_per_second": 13.146443681656958,
-    "gpu_batches_per_second": 1.0263281435977694,
-    "producer_headroom": 12.809201193267882,
+    "producer_batch_seconds": 0.05316818744530849,
+    "producer_p95_seconds": 0.1766920609989029,
+    "gpu_update_seconds": 0.9734568540102666,
+    "producer_batches_per_second": 18.808239438830803,
+    "gpu_batches_per_second": 1.027266895168888,
+    "producer_headroom": 18.30900959359606,
     "sustainable": true,
-    "buffer_seconds": 60.409529239508565,
+    "buffer_seconds": 60.35432494863653,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361588.754084028,
-    "upload_bytes_per_second": 5791522.386172582,
+    "generation_bytes_per_second": 361396.0339606937,
+    "upload_bytes_per_second": 5754149.303409202,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.58852927103644,
+    "estimated_full_shards_per_hour": 77.54717601886375,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 2663601,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.182102421998934,
+      "bytes": 5962190,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -24,3 +24,4 @@
 {"step": 282600, "loss": 0.6067937482148409, "lr": 0.0003, "grad_norm": 0.15201176093485552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41364822489, "tokens_per_second": 151013.7467797113, "data_wait_s": 1.795800017134752e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282700, "loss": 0.589450684376061, "lr": 0.0003, "grad_norm": 0.15634303942668046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41379596319, "tokens_per_second": 151108.18336983526, "data_wait_s": 1.4613999155699275e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282800, "loss": 0.653716541826725, "lr": 0.0003, "grad_norm": 0.1598144832022321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41394340143, "tokens_per_second": 150733.54300563177, "data_wait_s": 1.3720000424655154e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 282900, "loss": 0.5886063631623983, "lr": 0.0003, "grad_norm": 0.1679207484828719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41408818163, "tokens_per_second": 147980.88940932546, "data_wait_s": 8.70500116434414e-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:a4309349dd01f851788f2579c09b05d149f4aeb517223f1476f106bb8de0c546
-size 55233
+oid sha256:182c91b1119310cc7440291cb8884eb38c844e5f066daa0f7706f0bc517c63b8
+size 57351