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": 77488.5880998427,
-      "required_pages_per_second_upper_estimate": 0.005360840323042881
+      "required_token_positions_per_second": 77104.57245533603,
+      "required_pages_per_second_upper_estimate": 0.005334273229716853
     },
     {
       "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": 9686.073512480338,
-      "required_pages_per_second_upper_estimate": 0.004055356711459891
+      "required_token_positions_per_second": 9638.071556917004,
+      "required_pages_per_second_upper_estimate": 0.004035259295060342
     },
     {
       "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": 19372.147024960675,
-      "required_pages_per_second_upper_estimate": 0.004034707347427211
+      "required_token_positions_per_second": 19276.14311383401,
+      "required_pages_per_second_upper_estimate": 0.004014712264532919
     },
     {
       "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": 29058.220537441015,
-      "required_pages_per_second_upper_estimate": 0.001741995951885896
+      "required_token_positions_per_second": 28914.214670751015,
+      "required_pages_per_second_upper_estimate": 0.0017333630200620572
     },
     {
       "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": 29058.220537441015,
-      "required_pages_per_second_upper_estimate": 0.002259148833181653
+      "required_token_positions_per_second": 28914.214670751015,
+      "required_pages_per_second_upper_estimate": 0.0022479530104613714
     },
     {
       "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": 29058.220537441015,
-      "required_pages_per_second_upper_estimate": 0.37272765276793546
+      "required_token_positions_per_second": 28914.214670751015,
+      "required_pages_per_second_upper_estimate": 0.370880500131489
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.32388174496861666,
-    "producer_p95_seconds": 0.31020041200099513,
-    "gpu_update_seconds": 0.9514691363972556,
-    "producer_batches_per_second": 3.0875466602691595,
-    "gpu_batches_per_second": 1.051006240503509,
-    "producer_headroom": 2.937705354432528,
+    "producer_batch_seconds": 0.08915101277352733,
+    "producer_p95_seconds": 0.26406278500144253,
+    "gpu_update_seconds": 0.9562078830371316,
+    "producer_batches_per_second": 11.216922487917511,
+    "gpu_batches_per_second": 1.0457976949779737,
+    "producer_headroom": 10.7257097063632,
     "sustainable": true,
-    "buffer_seconds": 60.89402472942436,
+    "buffer_seconds": 60.24109663133929,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458187.6926747766,
-    "upload_bytes_per_second": 3886870.9094786476,
+    "generation_bytes_per_second": 459032.7444049806,
+    "upload_bytes_per_second": 3805509.369803716,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.31641278440927,
+    "estimated_full_shards_per_hour": 98.49774121391357,
     "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": 0.9234551270019438,
-      "bytes": 5872826,
-      "pages": 30,
-      "shards": 2
-    },
-    {
-      "seconds": 1.268712075001531,
-      "bytes": 10759229,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2720383849991777,
       "bytes": 7578314,
@@ -562,6 +550,18 @@
       "bytes": 10453891,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9155222249974031,
+      "bytes": 2430144,
+      "pages": 31,
+      "shards": 2
+    },
+    {
+      "seconds": 1.411119290998613,
+      "bytes": 5478989,
+      "pages": 19,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -248,3 +248,4 @@
 {"step": 549800, "loss": 0.5712734460830688, "lr": 0.0003, "grad_norm": 0.17033618191826674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80473228994, "tokens_per_second": 153308.82749478772, "data_wait_s": 1.3146000128472224e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549900, "loss": 0.6541333794593811, "lr": 0.0003, "grad_norm": 0.16892563216561565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80487929710, "tokens_per_second": 153447.29287662983, "data_wait_s": 9.311999747296795e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 550000, "loss": 0.6032522320747375, "lr": 0.0003, "grad_norm": 0.1624811246956349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80502732384, "tokens_per_second": 154663.50996704889, "data_wait_s": 9.3039998319e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 550100, "loss": 0.5845863819122314, "lr": 0.0003, "grad_norm": 0.1454932906480027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80517419446, "tokens_per_second": 133977.610676426, "data_wait_s": 1.0357001883676276e-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 sha256:142621ee14269f22c576c96cc179051a3c61a6c65b6ddfa9b94da9f88d48f8ab
-size 533580
+oid sha256:ff1652fd383180f31f812ecea50973d088dcf39b65be02fa485579705a53c5e7
+size 535698