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": 77391.17370114118,
-      "required_pages_per_second_upper_estimate": 0.0053541009689081626
+      "required_token_positions_per_second": 76862.99111059541,
+      "required_pages_per_second_upper_estimate": 0.005317560071741749
     },
     {
       "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": 9673.896712642647,
-      "required_pages_per_second_upper_estimate": 0.004050258539648346
+      "required_token_positions_per_second": 9607.873888824426,
+      "required_pages_per_second_upper_estimate": 0.004022616162028997
     },
     {
       "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": 19347.793425285294,
-      "required_pages_per_second_upper_estimate": 0.0040296351348624426
+      "required_token_positions_per_second": 19215.747777648852,
+      "required_pages_per_second_upper_estimate": 0.004002133508737672
     },
     {
       "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": 29021.69013792794,
-      "required_pages_per_second_upper_estimate": 0.0017398060102137775
+      "required_token_positions_per_second": 28823.621666473275,
+      "required_pages_per_second_upper_estimate": 0.0017279321077831146
     },
     {
       "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": 29021.69013792794,
-      "required_pages_per_second_upper_estimate": 0.0022563087552997575
+      "required_token_positions_per_second": 28823.621666473275,
+      "required_pages_per_second_upper_estimate": 0.0022409098028552906
     },
     {
       "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": 29021.69013792794,
-      "required_pages_per_second_upper_estimate": 0.3722590800262688
+      "required_token_positions_per_second": 28823.621666473275,
+      "required_pages_per_second_upper_estimate": 0.3697184703438036
     }
   ],
   "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": 0.08654564407046905,
-    "producer_p95_seconds": 0.24805656699754763,
-    "gpu_update_seconds": 0.952666776765951,
-    "producer_batches_per_second": 11.554596545445529,
-    "gpu_batches_per_second": 1.0496849731600095,
-    "producer_headroom": 11.007680247780584,
+    "producer_batch_seconds": 0.09338932092967411,
+    "producer_p95_seconds": 0.26305892799791764,
+    "gpu_update_seconds": 0.9592132564021015,
+    "producer_batches_per_second": 10.707862419869612,
+    "gpu_batches_per_second": 1.0425210382839003,
+    "producer_headroom": 10.271123580868817,
     "sustainable": true,
-    "buffer_seconds": 60.01800693625491,
+    "buffer_seconds": 60.4304351533324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288539.12747011567,
-    "upload_bytes_per_second": 4897488.826979845,
+    "generation_bytes_per_second": 288077.4260288065,
+    "upload_bytes_per_second": 4923207.640748212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.913779907966635,
+    "estimated_full_shards_per_hour": 61.81470952652116,
     "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": 0.9172461540001677,
-      "bytes": 2354642,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1485085579988663,
       "bytes": 12167168,
@@ -562,6 +556,12 @@
       "bytes": 7212495,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.216828146003536,
+      "bytes": 6096262,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -215,3 +215,4 @@
 {"step": 781500, "loss": 0.5546749234199524, "lr": 3e-05, "grad_norm": 0.2590823139928074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114389535494, "tokens_per_second": 152849.77736903654, "data_wait_s": 9.603001672076061e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 781600, "loss": 0.5506124496459961, "lr": 3e-05, "grad_norm": 0.24723907474109663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114404218556, "tokens_per_second": 153083.56295545545, "data_wait_s": 1.0194999049417675e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 781700, "loss": 0.505558967590332, "lr": 3e-05, "grad_norm": 0.2599157889450317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114418781227, "tokens_per_second": 150961.73112622736, "data_wait_s": 1.7187998309964314e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 781800, "loss": 0.5278440117835999, "lr": 3e-05, "grad_norm": 0.25320607190757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114433344427, "tokens_per_second": 150924.87152698849, "data_wait_s": 1.3950000720797107e-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:e3f53754fbd6cba0350db9c3f355e14717982007b210b2c70a76531c6bee1c8e
-size 462903
+oid sha256:7ca16301302387eb29f39448a7d619f632ecb8b3e2106826d568afef49131573
+size 465021