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": 76942.25922787549,
-      "required_pages_per_second_upper_estimate": 0.005323044024022557
+      "required_token_positions_per_second": 76440.30826564757,
+      "required_pages_per_second_upper_estimate": 0.005288317891768935
     },
     {
       "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": 9617.782403484436,
-      "required_pages_per_second_upper_estimate": 0.004026764650203828
+      "required_token_positions_per_second": 9555.038533205947,
+      "required_pages_per_second_upper_estimate": 0.004000495101959228
     },
     {
       "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": 19235.564806968872,
-      "required_pages_per_second_upper_estimate": 0.004006260873334826
+      "required_token_positions_per_second": 19110.077066411894,
+      "required_pages_per_second_upper_estimate": 0.003980125086311069
     },
     {
       "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": 28853.347210453307,
-      "required_pages_per_second_upper_estimate": 0.00172971410875644
+      "required_token_positions_per_second": 28665.115599617842,
+      "required_pages_per_second_upper_estimate": 0.001718429911099882
     },
     {
       "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": 28853.347210453307,
-      "required_pages_per_second_upper_estimate": 0.0022432208331508883
+      "required_token_positions_per_second": 28665.115599617842,
+      "required_pages_per_second_upper_estimate": 0.002228586653351787
     },
     {
       "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": 28853.347210453307,
-      "required_pages_per_second_upper_estimate": 0.37009975770517706
+      "required_token_positions_per_second": 28665.115599617842,
+      "required_pages_per_second_upper_estimate": 0.36768532470873694
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08518336581272479,
-    "producer_p95_seconds": 0.2713440530023945,
-    "gpu_update_seconds": 0.958225047455963,
-    "producer_batches_per_second": 11.739381162731878,
-    "gpu_batches_per_second": 1.0435961809336411,
-    "producer_headroom": 11.248969071762392,
+    "producer_batch_seconds": 0.08673439620332601,
+    "producer_p95_seconds": 0.2463992650009459,
+    "gpu_update_seconds": 0.9645173034072326,
+    "producer_batches_per_second": 11.529451333883303,
+    "gpu_batches_per_second": 1.0367880352871037,
+    "producer_headroom": 11.120355310322045,
     "sustainable": true,
-    "buffer_seconds": 60.36817798972567,
+    "buffer_seconds": 60.764590114655654,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 237498.7448770881,
-    "upload_bytes_per_second": 3973102.9977444257,
+    "generation_bytes_per_second": 264507.53591311467,
+    "upload_bytes_per_second": 3995250.4714760534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.96170196279986,
+    "estimated_full_shards_per_hour": 56.757159786654285,
     "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": 1.6904037299991614,
-      "bytes": 12097177,
-      "pages": 34,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2079874939990987,
-      "bytes": 1225243,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.7503577689985832,
       "bytes": 11722749,
@@ -562,6 +550,18 @@
       "bytes": 2119215,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.5652879460030817,
+      "bytes": 11222900,
+      "pages": 47,
+      "shards": 3
+    },
+    {
+      "seconds": 1.6341935630007356,
+      "bytes": 5648901,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -198,3 +198,4 @@
 {"step": 819800, "loss": 0.5151752233505249, "lr": 3e-05, "grad_norm": 0.2876552129358285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119995873760, "tokens_per_second": 150602.7573715653, "data_wait_s": 1.0370000381954014e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 819900, "loss": 0.4753550589084625, "lr": 3e-05, "grad_norm": 0.27392932714879936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120010509776, "tokens_per_second": 136167.57794326358, "data_wait_s": 1.544699989608489e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 820000, "loss": 0.5652066469192505, "lr": 3e-05, "grad_norm": 0.30110244378129664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120025135406, "tokens_per_second": 150709.33447864646, "data_wait_s": 1.1812000593636185e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 820100, "loss": 0.5288456082344055, "lr": 3e-05, "grad_norm": 0.28589186813386425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120039922242, "tokens_per_second": 141379.03867025502, "data_wait_s": 1.090099976863712e-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 sha256:99d4136f10c879216d5f1eb73284a6a460cb280f7357ac9772425e9b0a611d22
-size 426897
+oid sha256:4cf3146d5031caaa0ce5510bc07b81494898a1d31cf05be1a1cdf656bd6ee937
+size 429015