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": 76351.6878565838,
-      "required_pages_per_second_upper_estimate": 0.005282186926242224
+      "required_token_positions_per_second": 76732.8527363651,
+      "required_pages_per_second_upper_estimate": 0.005308556797049865
     },
     {
       "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": 9543.960982072975,
-      "required_pages_per_second_upper_estimate": 0.0039958571626254255
+      "required_token_positions_per_second": 9591.606592045637,
+      "required_pages_per_second_upper_estimate": 0.004015805384567503
     },
     {
       "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": 19087.92196414595,
-      "required_pages_per_second_upper_estimate": 0.00397551076277841
+      "required_token_positions_per_second": 19183.213184091273,
+      "required_pages_per_second_upper_estimate": 0.0039953574108945586
     },
     {
       "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": 28631.882946218924,
-      "required_pages_per_second_upper_estimate": 0.0017164376642719405
+      "required_token_positions_per_second": 28774.819776136912,
+      "required_pages_per_second_upper_estimate": 0.0017250065089736218
     },
     {
       "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": 28631.882946218924,
-      "required_pages_per_second_upper_estimate": 0.0022260029607250155
+      "required_token_positions_per_second": 28774.819776136912,
+      "required_pages_per_second_upper_estimate": 0.0022371156705384685
     },
     {
       "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": 28631.882946218924,
-      "required_pages_per_second_upper_estimate": 0.3672590519133788
+      "required_token_positions_per_second": 28774.819776136912,
+      "required_pages_per_second_upper_estimate": 0.3690924920939561
     }
   ],
   "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": 0.09438297439845655,
-    "producer_p95_seconds": 0.256755233000149,
-    "gpu_update_seconds": 0.9656368060715038,
-    "producer_batches_per_second": 10.595131233926795,
-    "gpu_batches_per_second": 1.0355860440617377,
-    "producer_headroom": 10.2310486846375,
+    "producer_batch_seconds": 0.11761466692976796,
+    "producer_p95_seconds": 0.279420871000184,
+    "gpu_update_seconds": 0.9608400752844545,
+    "producer_batches_per_second": 8.502340958863037,
+    "gpu_batches_per_second": 1.040755923616063,
+    "producer_headroom": 8.16938992700806,
     "sustainable": true,
-    "buffer_seconds": 60.83511878250474,
+    "buffer_seconds": 60.532924742920635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462481.8132335713,
-    "upload_bytes_per_second": 3236751.7556575267,
+    "generation_bytes_per_second": 445470.21626665426,
+    "upload_bytes_per_second": 3232139.665627795,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.23783109431605,
+    "estimated_full_shards_per_hour": 95.58753839492532,
     "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": 1.5439765690025524,
-      "bytes": 6266973,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.459664652000356,
       "bytes": 1024497,
@@ -562,6 +556,12 @@
       "bytes": 11759509,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.7052124439978797,
+      "bytes": 6177071,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 826400, "loss": 0.595355749130249, "lr": 3e-05, "grad_norm": 0.31233951201319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120961849689, "tokens_per_second": 152516.48072880716, "data_wait_s": 1.5437999536516145e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 826500, "loss": 0.4989027678966522, "lr": 3e-05, "grad_norm": 0.2870659709402653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120976436591, "tokens_per_second": 151059.66930902377, "data_wait_s": 7.548002031398937e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 826600, "loss": 0.5394002199172974, "lr": 3e-05, "grad_norm": 0.29511374486312314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120991040871, "tokens_per_second": 150849.54035445536, "data_wait_s": 7.058999472064897e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 826700, "loss": 0.5147653222084045, "lr": 3e-05, "grad_norm": 0.27749348607910346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121005498214, "tokens_per_second": 148855.0562910351, "data_wait_s": 1.1134001397294924e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5fd837ab26971f266ea57ce0384ef331c11b93edd929d5ba4d272f590cf42d0
-size 567468
+oid sha256:c3118737d3f67030c7352348b865b812118436fd3d860a592439bf0e6860bcfe
+size 569586