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": 75902.04352390835,
-      "required_pages_per_second_upper_estimate": 0.006578996760681814
+      "required_token_positions_per_second": 75297.77205218941,
+      "required_pages_per_second_upper_estimate": 0.00652662004102526
     },
     {
       "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": 9487.755440488543,
-      "required_pages_per_second_upper_estimate": 0.0037848074997959722
+      "required_token_positions_per_second": 9412.221506523676,
+      "required_pages_per_second_upper_estimate": 0.0037546758842044343
     },
     {
       "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": 18975.510880977086,
-      "required_pages_per_second_upper_estimate": 0.0043896504847286116
+      "required_token_positions_per_second": 18824.443013047352,
+      "required_pages_per_second_upper_estimate": 0.004354703592186737
     },
     {
       "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": 28463.26632146563,
-      "required_pages_per_second_upper_estimate": 0.001733263583760102
+      "required_token_positions_per_second": 28236.66451957103,
+      "required_pages_per_second_upper_estimate": 0.001719464722912493
     },
     {
       "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": 28463.26632146563,
-      "required_pages_per_second_upper_estimate": 0.002191181032790474
+      "required_token_positions_per_second": 28236.66451957103,
+      "required_pages_per_second_upper_estimate": 0.0021737365988066925
     },
     {
       "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": 28463.26632146563,
-      "required_pages_per_second_upper_estimate": 0.3534185072881487
+      "required_token_positions_per_second": 28236.66451957103,
+      "required_pages_per_second_upper_estimate": 0.35060487129606305
     }
   ],
   "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": 0.07584774253123072,
-    "producer_p95_seconds": 0.18442517700077588,
-    "gpu_update_seconds": 0.9713572464854184,
-    "producer_batches_per_second": 13.184308017977523,
-    "gpu_batches_per_second": 1.029487352483566,
-    "producer_headroom": 12.806673133158272,
+    "producer_batch_seconds": 0.07544566752350335,
+    "producer_p95_seconds": 0.21303432900003827,
+    "gpu_update_seconds": 0.97915247676782,
+    "producer_batches_per_second": 13.254571572164474,
+    "gpu_batches_per_second": 1.0212913961071697,
+    "producer_headroom": 12.978246583381184,
     "sustainable": true,
-    "buffer_seconds": 60.224149282095944,
+    "buffer_seconds": 60.70745355960484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345021.53710020595,
-    "upload_bytes_per_second": 3559192.2932783407,
+    "generation_bytes_per_second": 312153.2704713088,
+    "upload_bytes_per_second": 3612803.896304032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.03359017138132,
+    "estimated_full_shards_per_hour": 66.98082528690765,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 2343472,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3791698659997564,
+      "bytes": 10729274,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 244200, "loss": 0.5939223114401102, "lr": 0.0003, "grad_norm": 0.1427175928723141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35743579350, "tokens_per_second": 148211.79195562232, "data_wait_s": 1.0879000001295935e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 244300, "loss": 0.6783288093283772, "lr": 0.0003, "grad_norm": 0.15650033404221986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35758060279, "tokens_per_second": 147907.7248723915, "data_wait_s": 1.6136000340338796e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 244400, "loss": 0.6831602342426777, "lr": 0.0003, "grad_norm": 0.19541405455019706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35772676862, "tokens_per_second": 138980.74329869053, "data_wait_s": 1.78360005520517e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 244500, "loss": 0.5546451276168227, "lr": 0.0003, "grad_norm": 0.15389539280029582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35787392726, "tokens_per_second": 149817.3266908133, "data_wait_s": 1.715200050966814e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2c32cc1f97a69d2dc22148e056b8389d2d408431209e67dbea911298b2873b0
-size 155562
+oid sha256:d043c2c35c1a5c7580421e1de72002336de118bc7128f88d11f06fecf5ea2845
+size 157680