Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77488.57724680714,
-      "required_pages_per_second_upper_estimate": 0.005360839572204698
+      "required_token_positions_per_second": 77519.73779852093,
+      "required_pages_per_second_upper_estimate": 0.0053629953314747
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9686.072155850892,
-      "required_pages_per_second_upper_estimate": 0.004055356143467472
+      "required_token_positions_per_second": 9689.967224815116,
+      "required_pages_per_second_upper_estimate": 0.004056986927504503
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19372.144311701784,
-      "required_pages_per_second_upper_estimate": 0.004034706782326938
+      "required_token_positions_per_second": 19379.934449630233,
+      "required_pages_per_second_upper_estimate": 0.004036329262617681
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29058.216467552676,
-      "required_pages_per_second_upper_estimate": 0.0017419957079023044
+      "required_token_positions_per_second": 29069.90167444535,
+      "required_pages_per_second_upper_estimate": 0.0017426962182132424
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29058.216467552676,
-      "required_pages_per_second_upper_estimate": 0.002259148516765736
+      "required_token_positions_per_second": 29069.90167444535,
+      "required_pages_per_second_upper_estimate": 0.0022600569902038485
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29058.216467552676,
-      "required_pages_per_second_upper_estimate": 0.37272760056377774
+      "required_token_positions_per_second": 29069.90167444535,
+      "required_pages_per_second_upper_estimate": 0.37287748585119934
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4747609925468339,
-    "producer_p95_seconds": 0.23850065099759377,
-    "gpu_update_seconds": 0.9514692696598441,
-    "producer_batches_per_second": 2.106323003993115,
-    "gpu_batches_per_second": 1.0510060932997929,
-    "producer_headroom": 2.004101610277058,
+    "producer_batch_seconds": 0.09904365101564849,
+    "producer_p95_seconds": 0.2272852749993035,
+    "gpu_update_seconds": 0.9510868082606792,
+    "producer_batches_per_second": 10.0965583330728,
+    "gpu_batches_per_second": 1.0514287353315013,
+    "producer_headroom": 9.602703439419972,
     "sustainable": true,
-    "buffer_seconds": 60.894033258230024,
+    "buffer_seconds": 60.869555728683466,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335712.00863941165,
-    "upload_bytes_per_second": 5819690.783530984,
+    "generation_bytes_per_second": 335482.34090936073,
+    "upload_bytes_per_second": 5687604.329029913,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.035982078426,
+    "estimated_full_shards_per_hour": 71.9867007299482,
     "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": 1.5313736190000782,
-      "bytes": 6009990,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0109308320006676,
       "bytes": 5084502,
@@ -562,6 +556,12 @@
       "bytes": 6513685,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 4.352831413998501,
+      "bytes": 12282177,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -156,3 +156,4 @@
 {"step": 500600, "loss": 0.6207348108291626, "lr": 0.0003, "grad_norm": 0.17324735857896328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73270893517, "tokens_per_second": 151284.24541172347, "data_wait_s": 9.136001608567312e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 500700, "loss": 0.5542911291122437, "lr": 0.0003, "grad_norm": 0.1595898970018304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73285544209, "tokens_per_second": 143865.43909430673, "data_wait_s": 9.431001672055572e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 500800, "loss": 0.5830085873603821, "lr": 0.0003, "grad_norm": 0.1689505062686292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73300241182, "tokens_per_second": 153446.22490661117, "data_wait_s": 1.5358000382548198e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 500900, "loss": 0.5516468286514282, "lr": 0.0003, "grad_norm": 0.1395681611502911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73314884916, "tokens_per_second": 152671.84020874542, "data_wait_s": 1.1242998880334198e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:468abcd9854e6274982e007544ffc64b8f1333f06cab04c1aaec43f34c3c2302
-size 337158
+oid sha256:f2f71c20ca4d9dbd453dc82de10c702b91cbf86140c2c55198dd4bd874598f85
+size 339276