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": 76942.14789044435,
-      "required_pages_per_second_upper_estimate": 0.005323036321440732
+      "required_token_positions_per_second": 77079.53862303727,
+      "required_pages_per_second_upper_estimate": 0.005332541331111931
     },
     {
       "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": 9617.768486305544,
-      "required_pages_per_second_upper_estimate": 0.00402675882337165
+      "required_token_positions_per_second": 9634.942327879658,
+      "required_pages_per_second_upper_estimate": 0.004033949152208138
     },
     {
       "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": 19235.536972611088,
-      "required_pages_per_second_upper_estimate": 0.004006255076172141
+      "required_token_positions_per_second": 19269.884655759317,
+      "required_pages_per_second_upper_estimate": 0.004013408792762599
     },
     {
       "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": 28853.305458916635,
-      "required_pages_per_second_upper_estimate": 0.0017297116058155675
+      "required_token_positions_per_second": 28904.826983638977,
+      "required_pages_per_second_upper_estimate": 0.0017328002425538588
     },
     {
       "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": 28853.305458916635,
-      "required_pages_per_second_upper_estimate": 0.0022432175871525575
+      "required_token_positions_per_second": 28904.826983638977,
+      "required_pages_per_second_upper_estimate": 0.0022472231590804823
     },
     {
       "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": 28853.305458916635,
-      "required_pages_per_second_upper_estimate": 0.37009922216129393
+      "required_token_positions_per_second": 28904.826983638977,
+      "required_pages_per_second_upper_estimate": 0.3707600849609289
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.42406271990617483,
-    "producer_p95_seconds": 0.27598853199742734,
-    "gpu_update_seconds": 0.958226434034297,
-    "producer_batches_per_second": 2.3581417395550663,
-    "gpu_batches_per_second": 1.0435946708230843,
-    "producer_headroom": 2.259633750041285,
+    "producer_batch_seconds": 0.09390799681256112,
+    "producer_p95_seconds": 0.26356070499969064,
+    "gpu_update_seconds": 0.95651843948589,
+    "producer_batches_per_second": 10.648720385293535,
+    "gpu_batches_per_second": 1.0454581518966644,
+    "producer_headroom": 10.185697405462557,
     "sustainable": true,
-    "buffer_seconds": 60.36826534416071,
+    "buffer_seconds": 60.26066168761107,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368545.6908781735,
-    "upload_bytes_per_second": 3807510.4743306492,
+    "generation_bytes_per_second": 354354.3781267119,
+    "upload_bytes_per_second": 3902259.7763395305,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.08132595786003,
+    "estimated_full_shards_per_hour": 76.0362005982496,
     "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": 1.0708622599995579,
-      "bytes": 1033241,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.2405112279993773,
       "bytes": 12607883,
@@ -562,6 +556,12 @@
       "bytes": 6693523,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1304304789991875,
+      "bytes": 12005280,
+      "pages": 46,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -267,3 +267,4 @@
 {"step": 551700, "loss": 0.5444007515907288, "lr": 0.0003, "grad_norm": 0.14396969298588103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80751490379, "tokens_per_second": 151359.89025243273, "data_wait_s": 1.1003998224623501e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551800, "loss": 0.5731511116027832, "lr": 0.0003, "grad_norm": 0.14905357860602683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80766091810, "tokens_per_second": 152191.24941692536, "data_wait_s": 1.7607999325264245e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551900, "loss": 0.5963700413703918, "lr": 0.0003, "grad_norm": 0.15125072113458699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80780646165, "tokens_per_second": 151248.49202182028, "data_wait_s": 8.867002179613337e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 552000, "loss": 0.611603856086731, "lr": 0.0003, "grad_norm": 0.1777720764578613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80795224140, "tokens_per_second": 152007.90497875676, "data_wait_s": 2.2330998035613447e-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 sha256:2567c82cc0354dc9f037bf77a72d333ee48577ea173cf66dedb3ba30b954000e
-size 573822
+oid sha256:86881ed8112c43c406ab0de891023c109799f3e2c2accef02abc459473118ec6
+size 575940