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": 76771.20410586511,
-      "required_pages_per_second_upper_estimate": 0.005311210033779313
+      "required_token_positions_per_second": 77496.96629334897,
+      "required_pages_per_second_upper_estimate": 0.005361419945909732
     },
     {
       "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": 9596.400513233139,
-      "required_pages_per_second_upper_estimate": 0.004017812499260252
+      "required_token_positions_per_second": 9687.120786668622,
+      "required_pages_per_second_upper_estimate": 0.004055795183292954
     },
     {
       "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": 19192.801026466277,
-      "required_pages_per_second_upper_estimate": 0.003997354305612866
+      "required_token_positions_per_second": 19374.241573337244,
+      "required_pages_per_second_upper_estimate": 0.004035143586617096
     },
     {
       "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": 28789.201539699417,
-      "required_pages_per_second_upper_estimate": 0.0017258686737395125
+      "required_token_positions_per_second": 29061.362360005864,
+      "required_pages_per_second_upper_estimate": 0.0017421842993513733
     },
     {
       "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": 28789.201539699417,
-      "required_pages_per_second_upper_estimate": 0.0022382337893967607
+      "required_token_positions_per_second": 29061.362360005864,
+      "required_pages_per_second_upper_estimate": 0.0022593930960666526
     },
     {
       "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": 28789.201539699417,
-      "required_pages_per_second_upper_estimate": 0.3692769659150013
+      "required_token_positions_per_second": 29061.362360005864,
+      "required_pages_per_second_upper_estimate": 0.3727679526943711
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0835966003513704,
-    "producer_p95_seconds": 0.22870677799801342,
-    "gpu_update_seconds": 0.9603600836888186,
-    "producer_batches_per_second": 11.962208939081659,
-    "gpu_batches_per_second": 1.0412760973560264,
-    "producer_headroom": 11.488027977839597,
+    "producer_batch_seconds": 0.06409569122646985,
+    "producer_p95_seconds": 0.22019659800207592,
+    "gpu_update_seconds": 0.951366273112133,
+    "producer_batches_per_second": 15.601672762474648,
+    "gpu_batches_per_second": 1.051119877025675,
+    "producer_headroom": 14.842905270350583,
     "sustainable": true,
-    "buffer_seconds": 60.502685272395574,
+    "buffer_seconds": 60.88744147917651,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329119.28774242365,
-    "upload_bytes_per_second": 5524789.107582198,
+    "generation_bytes_per_second": 328486.33021315566,
+    "upload_bytes_per_second": 5487822.071607167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.62133764462025,
+    "estimated_full_shards_per_hour": 70.48551969333651,
     "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": 1.2196053769985156,
-      "bytes": 6057143,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2347321990000637,
       "bytes": 9643472,
@@ -562,6 +556,12 @@
       "bytes": 12124317,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8537721579996287,
+      "bytes": 1128490,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 545700, "loss": 0.5606106519699097, "lr": 0.0003, "grad_norm": 0.1571157613446879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79873456688, "tokens_per_second": 139604.44434697292, "data_wait_s": 1.5087000065250322e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 545800, "loss": 0.627779483795166, "lr": 0.0003, "grad_norm": 0.21038275067424334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79888209184, "tokens_per_second": 153975.8169430534, "data_wait_s": 1.6963000234682113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 545900, "loss": 0.5648422241210938, "lr": 0.0003, "grad_norm": 0.1479416133957257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79902905568, "tokens_per_second": 153692.59561944264, "data_wait_s": 1.0698000551201403e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 546000, "loss": 0.5812898874282837, "lr": 0.0003, "grad_norm": 0.20790069483774973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79917449390, "tokens_per_second": 151440.93368076577, "data_wait_s": 1.1511998309288174e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c983f0ec9171efbec0bf5b3b7b069a5911fa6dc7aff6d258f0ab8d04c657bef
-size 445959
+oid sha256:288abd03598f1d596b691cb35f4525d496219a6bc8a25b18cef0c0bff3c052a3
+size 448077