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": 75745.6130420059,
-      "required_pages_per_second_upper_estimate": 0.006565437762979943
+      "required_token_positions_per_second": 75565.5466849141,
+      "required_pages_per_second_upper_estimate": 0.006549830067521234
     },
     {
       "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": 9468.201630250738,
-      "required_pages_per_second_upper_estimate": 0.003777007192536596
+      "required_token_positions_per_second": 9445.693335614262,
+      "required_pages_per_second_upper_estimate": 0.003768028297277111
     },
     {
       "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": 18936.403260501476,
-      "required_pages_per_second_upper_estimate": 0.004380603625002192
+      "required_token_positions_per_second": 18891.386671228523,
+      "required_pages_per_second_upper_estimate": 0.004370189829338807
     },
     {
       "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": 28404.604890752213,
-      "required_pages_per_second_upper_estimate": 0.0017296914104023957
+      "required_token_positions_per_second": 28337.080006842785,
+      "required_pages_per_second_upper_estimate": 0.0017255794992479976
     },
     {
       "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": 28404.604890752213,
-      "required_pages_per_second_upper_estimate": 0.0021866651134689213
+      "required_token_positions_per_second": 28337.080006842785,
+      "required_pages_per_second_upper_estimate": 0.002181466860984732
     },
     {
       "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": 28404.604890752213,
-      "required_pages_per_second_upper_estimate": 0.3526901286458673
+      "required_token_positions_per_second": 28337.080006842785,
+      "required_pages_per_second_upper_estimate": 0.35185169557896107
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.4270988903281534,
-    "producer_p95_seconds": 0.34832001500035403,
-    "gpu_update_seconds": 0.973363301701882,
-    "producer_batches_per_second": 2.341378127280239,
-    "gpu_batches_per_second": 1.027365628282415,
-    "producer_headroom": 2.2790115445020627,
+    "producer_batch_seconds": 0.11425002277351837,
+    "producer_p95_seconds": 0.29003522800121573,
+    "gpu_update_seconds": 0.9756827447755242,
+    "producer_batches_per_second": 8.752733485071888,
+    "gpu_batches_per_second": 1.024923322006756,
+    "producer_headroom": 8.53989103100358,
     "sustainable": true,
-    "buffer_seconds": 60.34852470551669,
+    "buffer_seconds": 60.4923301760825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364468.8788837926,
-    "upload_bytes_per_second": 3319725.431627655,
+    "generation_bytes_per_second": 376750.64256922714,
+    "upload_bytes_per_second": 3416057.1548103867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.20653700719198,
+    "estimated_full_shards_per_hour": 80.8419175892602,
     "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": 2.9626744119996147,
-      "bytes": 6098299,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1153004340012558,
       "bytes": 6713918,
@@ -562,6 +556,12 @@
       "bytes": 6833428,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.242159831999743,
+      "bytes": 12337974,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -231,3 +231,4 @@
 {"step": 303300, "loss": 0.5642993152141571, "lr": 0.0003, "grad_norm": 0.15055635545494098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44395360099, "tokens_per_second": 149010.12785942582, "data_wait_s": 1.1391999578336254e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303400, "loss": 0.563585702329874, "lr": 0.0003, "grad_norm": 0.15217523672935845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44409861465, "tokens_per_second": 147336.13125779293, "data_wait_s": 1.5006000467110425e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303500, "loss": 0.5572625398635864, "lr": 0.0003, "grad_norm": 0.14108857983182388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44424505103, "tokens_per_second": 149892.5680040023, "data_wait_s": 1.1225998605368659e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 303600, "loss": 0.612817732617259, "lr": 0.0003, "grad_norm": 0.1585622809536886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44439193986, "tokens_per_second": 149973.40725128085, "data_wait_s": 7.378999725915492e-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 sha256:5f785f3919ba58e86f3a16f0f88f807bd43c9e9a604d9c5ae525d624d67d7993
-size 496791
+oid sha256:dc39cae368534f62d46f667e2e47eef97600c04ddb0fa01e9228743b4947baa5
+size 498909