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": 76589.24498588534,
-      "required_pages_per_second_upper_estimate": 0.0052986216796557855
+      "required_token_positions_per_second": 76887.42303925686,
+      "required_pages_per_second_upper_estimate": 0.005319250329256432
     },
     {
       "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": 9573.655623235667,
-      "required_pages_per_second_upper_estimate": 0.004008289688785624
+      "required_token_positions_per_second": 9610.927879907107,
+      "required_pages_per_second_upper_estimate": 0.00402389480432073
     },
     {
       "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": 19147.311246471334,
-      "required_pages_per_second_upper_estimate": 0.00398787998408609
+      "required_token_positions_per_second": 19221.855759814214,
+      "required_pages_per_second_upper_estimate": 0.004003405640344396
     },
     {
       "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": 28720.966869707,
-      "required_pages_per_second_upper_estimate": 0.0017217781094617756
+      "required_token_positions_per_second": 28832.78363972132,
+      "required_pages_per_second_upper_estimate": 0.001728481354089812
     },
     {
       "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": 28720.966869707,
-      "required_pages_per_second_upper_estimate": 0.0022329288439374446
+      "required_token_positions_per_second": 28832.78363972132,
+      "required_pages_per_second_upper_estimate": 0.002241622105976065
     },
     {
       "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": 28720.966869707,
-      "required_pages_per_second_upper_estimate": 0.3684017248330191
+      "required_token_positions_per_second": 28832.78363972132,
+      "required_pages_per_second_upper_estimate": 0.3698359902992691
     }
   ],
   "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": 0.10013397956231529,
-    "producer_p95_seconds": 0.27769846199953463,
-    "gpu_update_seconds": 0.962641687009545,
-    "producer_batches_per_second": 9.986619970273736,
-    "gpu_batches_per_second": 1.0388081188406757,
-    "producer_headroom": 9.613536695707522,
+    "producer_batch_seconds": 0.5140078114141318,
+    "producer_p95_seconds": 0.2751062269999238,
+    "gpu_update_seconds": 0.9589084545382184,
+    "producer_batches_per_second": 1.9454957255392922,
+    "gpu_batches_per_second": 1.0428524175246427,
+    "producer_headroom": 1.8655522994875926,
     "sustainable": true,
-    "buffer_seconds": 60.646426281601336,
+    "buffer_seconds": 60.41123263590776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288485.1748269605,
-    "upload_bytes_per_second": 5253353.013598868,
+    "generation_bytes_per_second": 262358.10019762133,
+    "upload_bytes_per_second": 5259035.961411815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.90220292669879,
+    "estimated_full_shards_per_hour": 56.295940918412015,
     "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": 1.3444492599992373,
-      "bytes": 6214787,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2648084209977242,
-      "bytes": 5986982,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.4115151900005003,
       "bytes": 7731251,
@@ -562,6 +550,18 @@
       "bytes": 6985484,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4114699799974915,
+      "bytes": 10209990,
+      "pages": 36,
+      "shards": 3
+    },
+    {
+      "seconds": 1.0779238770010124,
+      "bytes": 1827152,
+      "pages": 11,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 749800, "loss": 0.5130945444107056, "lr": 5e-05, "grad_norm": 0.22538138410973832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109749564598, "tokens_per_second": 152521.85043806894, "data_wait_s": 1.4295001165010035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 749900, "loss": 0.5240430235862732, "lr": 5e-05, "grad_norm": 0.2319403982393609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109764255929, "tokens_per_second": 150777.05198659722, "data_wait_s": 1.2971002433914691e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750000, "loss": 0.550243616104126, "lr": 5e-05, "grad_norm": 0.24450628622732434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109778926010, "tokens_per_second": 151990.02936366422, "data_wait_s": 1.3694996596314013e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 750100, "loss": 0.5139143466949463, "lr": 5e-05, "grad_norm": 0.23780783135987257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109793600219, "tokens_per_second": 139187.46416654414, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8b855552e8e53bee00787f8ffcda6efb211f6b2eb42a0fd993123937b517afa
-size 640263
+oid sha256:93ffee259eccafaf5a92a3bdb51c66f7197986b7c7503101b115b015613a4531
+size 642381