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": 77278.43130618232,
-      "required_pages_per_second_upper_estimate": 0.005346301188426512
+      "required_token_positions_per_second": 77171.23889603485,
+      "required_pages_per_second_upper_estimate": 0.005338885368771851
     },
     {
       "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": 9659.80391327279,
-      "required_pages_per_second_upper_estimate": 0.004044358178843303
+      "required_token_positions_per_second": 9646.404862004356,
+      "required_pages_per_second_upper_estimate": 0.004038748275881217
     },
     {
       "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": 19319.60782654558,
-      "required_pages_per_second_upper_estimate": 0.004023764817949184
+      "required_token_positions_per_second": 19292.80972400871,
+      "required_pages_per_second_upper_estimate": 0.004018183479904242
     },
     {
       "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": 28979.411739818366,
-      "required_pages_per_second_upper_estimate": 0.0017372714848024845
+      "required_token_positions_per_second": 28939.214586013066,
+      "required_pages_per_second_upper_estimate": 0.0017348617268093567
     },
     {
       "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": 28979.411739818366,
-      "required_pages_per_second_upper_estimate": 0.0022530217958097584
+      "required_token_positions_per_second": 28939.214586013066,
+      "required_pages_per_second_upper_estimate": 0.00224989664390999
     },
     {
       "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": 28979.411739818366,
-      "required_pages_per_second_upper_estimate": 0.3717167781303263
+      "required_token_positions_per_second": 28939.214586013066,
+      "required_pages_per_second_upper_estimate": 0.37120117220165294
     }
   ],
   "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": 0.12543926564842423,
-    "producer_p95_seconds": 0.3934989929985022,
-    "gpu_update_seconds": 0.9540566333170601,
-    "producer_batches_per_second": 7.971985445153649,
-    "gpu_batches_per_second": 1.0481558065617176,
-    "producer_headroom": 7.605725594655895,
+    "producer_batch_seconds": 0.12147098764046405,
+    "producer_p95_seconds": 0.37642398000025423,
+    "gpu_update_seconds": 0.9553818372583913,
+    "producer_batches_per_second": 8.232418451719932,
+    "gpu_batches_per_second": 1.0467019164501254,
+    "producer_headroom": 7.86510306548407,
     "sustainable": true,
-    "buffer_seconds": 60.10556789897479,
+    "buffer_seconds": 60.18905574727865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335917.32861683815,
-    "upload_bytes_per_second": 5017857.881858054,
+    "generation_bytes_per_second": 334279.07846238994,
+    "upload_bytes_per_second": 5011272.488277926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.08003896597727,
+    "estimated_full_shards_per_hour": 71.72850861934445,
     "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": 1.027337190000253,
-      "bytes": 5987324,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.2340771259987378,
       "bytes": 10655993,
@@ -562,6 +556,12 @@
       "bytes": 12259202,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.3878037089962163,
+      "bytes": 7215998,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 686000, "loss": 0.5922773480415344, "lr": 0.0001, "grad_norm": 0.2260045671226013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100412502302, "tokens_per_second": 152276.9905235283, "data_wait_s": 1.1070995242334902e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686100, "loss": 0.59559565782547, "lr": 0.0001, "grad_norm": 0.20903594979744555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100427191750, "tokens_per_second": 152045.6171866706, "data_wait_s": 1.3903998478781432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686200, "loss": 0.5211208462715149, "lr": 0.0001, "grad_norm": 0.21976697175349297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100441963613, "tokens_per_second": 153376.83984189905, "data_wait_s": 2.062300336547196e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 686300, "loss": 0.5422401428222656, "lr": 0.0001, "grad_norm": 0.2058293217477818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100456651841, "tokens_per_second": 152989.9583270223, "data_wait_s": 9.724004485178739e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a676dd9072d087ebd0c64301aed1c5c6db4279837124ba7ef1fd9ede13ca293
-size 879045
+oid sha256:e5e177b7827f114c505201b86f75750af7514bc6dc265101f976fa3449e1c19a
+size 881163