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": 77365.42010107003,
-      "required_pages_per_second_upper_estimate": 0.005352319275098656
+      "required_token_positions_per_second": 77410.70549822866,
+      "required_pages_per_second_upper_estimate": 0.005355452223950689
     },
     {
       "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": 9670.677512633754,
-      "required_pages_per_second_upper_estimate": 0.0040489107278291634
+      "required_token_positions_per_second": 9676.338187278583,
+      "required_pages_per_second_upper_estimate": 0.0040512807340946245
     },
     {
       "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": 19341.35502526751,
-      "required_pages_per_second_upper_estimate": 0.004028294185930764
+      "required_token_positions_per_second": 19352.676374557166,
+      "required_pages_per_second_upper_estimate": 0.004030652124423741
     },
     {
       "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": 29012.032537901265,
-      "required_pages_per_second_upper_estimate": 0.0017392270518384798
+      "required_token_positions_per_second": 29029.014561835753,
+      "required_pages_per_second_upper_estimate": 0.001740245098243303
     },
     {
       "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": 29012.032537901265,
-      "required_pages_per_second_upper_estimate": 0.0022555579193769765
+      "required_token_positions_per_second": 29029.014561835753,
+      "required_pages_per_second_upper_estimate": 0.002256878197041854
     },
     {
       "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": 29012.032537901265,
-      "required_pages_per_second_upper_estimate": 0.3721352027026496
+      "required_token_positions_per_second": 29029.014561835753,
+      "required_pages_per_second_upper_estimate": 0.3723530298718045
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07821037822658639,
-    "producer_p95_seconds": 0.2139383989997441,
-    "gpu_update_seconds": 0.9529839029334021,
-    "producer_batches_per_second": 12.786026901734964,
-    "gpu_batches_per_second": 1.049335667603489,
-    "producer_headroom": 12.18487781982686,
+    "producer_batch_seconds": 0.057982702656318,
+    "producer_p95_seconds": 0.20746183300070697,
+    "gpu_update_seconds": 0.952426405695102,
+    "producer_batches_per_second": 17.24652274191701,
+    "gpu_batches_per_second": 1.0499498901126936,
+    "producer_headroom": 16.42604366582285,
     "sustainable": true,
-    "buffer_seconds": 60.03798588480433,
+    "buffer_seconds": 60.002863558791425,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284199.13156234816,
-    "upload_bytes_per_second": 4217755.658593181,
+    "generation_bytes_per_second": 283178.41236008686,
+    "upload_bytes_per_second": 4220298.999030443,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.98251781609377,
+    "estimated_full_shards_per_hour": 60.76349523641543,
     "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": 1.1475876189997507,
-      "bytes": 6414179,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1066383229999701,
       "bytes": 6242311,
@@ -562,6 +556,12 @@
       "bytes": 6993608,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.010654464000254,
+      "bytes": 6096042,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 698800, "loss": 0.5014955401420593, "lr": 8e-05, "grad_norm": 0.19655496331937625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102285337793, "tokens_per_second": 140519.8405952182, "data_wait_s": 1.6682999557815492e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698900, "loss": 0.5420318841934204, "lr": 8e-05, "grad_norm": 0.21903995589622968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102299964981, "tokens_per_second": 152612.15655848893, "data_wait_s": 2.1781001123599708e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699000, "loss": 0.5323392748832703, "lr": 8e-05, "grad_norm": 0.20052000933499528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102314636869, "tokens_per_second": 152673.2889912569, "data_wait_s": 2.107599902956281e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 699100, "loss": 0.49232256412506104, "lr": 8e-05, "grad_norm": 0.21950936818792338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102329243862, "tokens_per_second": 152557.72502462505, "data_wait_s": 1.537200114398729e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97db78c38c1a1d2b8b6ce903a5756170a676cc0647b2d08bcea7e0d22bfca043
-size 298251
+oid sha256:91f381aa69fb6102ce63be85f52969b00abe737672220651a83768e0d115b116
+size 300369