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": 77244.17117457373,
-      "required_pages_per_second_upper_estimate": 0.0053439309930273715
+      "required_token_positions_per_second": 76774.05654670329,
+      "required_pages_per_second_upper_estimate": 0.005311407372255073
     },
     {
       "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": 9655.521396821716,
-      "required_pages_per_second_upper_estimate": 0.004042565178634351
+      "required_token_positions_per_second": 9596.75706833791,
+      "required_pages_per_second_upper_estimate": 0.004017961781436903
     },
     {
       "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": 19311.04279364343,
-      "required_pages_per_second_upper_estimate": 0.004021980947470773
+      "required_token_positions_per_second": 19193.51413667582,
+      "required_pages_per_second_upper_estimate": 0.003997502827663532
     },
     {
       "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": 28966.564190465146,
-      "required_pages_per_second_upper_estimate": 0.001736501293835831
+      "required_token_positions_per_second": 28790.271205013734,
+      "required_pages_per_second_upper_estimate": 0.0017259327985418213
     },
     {
       "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": 28966.564190465146,
-      "required_pages_per_second_upper_estimate": 0.0022520229553579427
+      "required_token_positions_per_second": 28790.271205013734,
+      "required_pages_per_second_upper_estimate": 0.0022383169511699884
     },
     {
       "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": 28966.564190465146,
-      "required_pages_per_second_upper_estimate": 0.3715519835618469
+      "required_token_positions_per_second": 28790.271205013734,
+      "required_pages_per_second_upper_estimate": 0.36929068643313623
     }
   ],
   "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": 0.12210022925785324,
-    "producer_p95_seconds": 0.38375506599550135,
-    "gpu_update_seconds": 0.954479786356603,
-    "producer_batches_per_second": 8.189992812283618,
-    "gpu_batches_per_second": 1.0476911237870785,
-    "producer_headroom": 7.817182589730582,
+    "producer_batch_seconds": 0.4091385062965287,
+    "producer_p95_seconds": 0.3623796640022192,
+    "gpu_update_seconds": 0.960324402751204,
+    "producer_batches_per_second": 2.4441600695370296,
+    "gpu_batches_per_second": 1.0413147860609713,
+    "producer_headroom": 2.3471865590064893,
     "sustainable": true,
-    "buffer_seconds": 60.13222654046599,
+    "buffer_seconds": 60.50043737332585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494987.90737296373,
-    "upload_bytes_per_second": 4517615.744659073,
+    "generation_bytes_per_second": 443325.19472720014,
+    "upload_bytes_per_second": 4487753.8827649625,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.21288219348607,
+    "estimated_full_shards_per_hour": 95.12726670610431,
     "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": 1.198663317001774,
-      "bytes": 9767212,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.4251030520026688,
       "bytes": 1167175,
@@ -562,6 +556,12 @@
       "bytes": 5878935,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.573873765002645,
+      "bytes": 8582970,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -437,3 +437,4 @@
 {"step": 803700, "loss": 0.5388901233673096, "lr": 3e-05, "grad_norm": 0.2735688570848663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117638367567, "tokens_per_second": 150629.6751526873, "data_wait_s": 9.963005140889436e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 803800, "loss": 0.5099372267723083, "lr": 3e-05, "grad_norm": 0.2714258411883778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117652931556, "tokens_per_second": 151436.96065108583, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803900, "loss": 0.5207173824310303, "lr": 3e-05, "grad_norm": 0.2727557766442316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117667628222, "tokens_per_second": 152070.79059758803, "data_wait_s": 9.208997653331608e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 804000, "loss": 0.555134654045105, "lr": 3e-05, "grad_norm": 0.2694330548798989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117682351354, "tokens_per_second": 152743.03588176318, "data_wait_s": 1.2853000953327864e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4e8f5248edb9506c96f9dbf706a224ad6712934470ba18f9284065d7237060d
-size 936231
+oid sha256:113fbae046f4d69401b79bee0939e4a1cb1eb6556739fc9219f306876262ed69
+size 938349