Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75293.69499126023,
-      "required_pages_per_second_upper_estimate": 0.006526266651717135
+      "required_token_positions_per_second": 75718.55996863873,
+      "required_pages_per_second_upper_estimate": 0.006563092871145863
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9411.711873907529,
-      "required_pages_per_second_upper_estimate": 0.0037544725841341664
+      "required_token_positions_per_second": 9464.819996079841,
+      "required_pages_per_second_upper_estimate": 0.0037756582081058883
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18823.423747815057,
-      "required_pages_per_second_upper_estimate": 0.004354467803113701
+      "required_token_positions_per_second": 18929.639992159682,
+      "required_pages_per_second_upper_estimate": 0.004379039061900776
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28235.135621722584,
-      "required_pages_per_second_upper_estimate": 0.0017193716210550298
+      "required_token_positions_per_second": 28394.459988239523,
+      "required_pages_per_second_upper_estimate": 0.0017290736390655629
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28235.135621722584,
-      "required_pages_per_second_upper_estimate": 0.0021736189000180588
+      "required_token_positions_per_second": 28394.459988239523,
+      "required_pages_per_second_upper_estimate": 0.002185884130790607
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28235.135621722584,
-      "required_pages_per_second_upper_estimate": 0.3505858875016773
+      "required_token_positions_per_second": 28394.459988239523,
+      "required_pages_per_second_upper_estimate": 0.35256416290946424
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12257916046871742,
-    "producer_p95_seconds": 0.2969676510001591,
-    "gpu_update_seconds": 0.9792054966695157,
-    "producer_batches_per_second": 8.157993546180332,
-    "gpu_batches_per_second": 1.0212360974292023,
-    "producer_headroom": 7.988352122214216,
+    "producer_batch_seconds": 0.1190939332032599,
+    "producer_p95_seconds": 0.23543296000207192,
+    "gpu_update_seconds": 0.9737110693935122,
+    "producer_batches_per_second": 8.39673334403425,
+    "gpu_batches_per_second": 1.0269986974913023,
+    "producer_headroom": 8.17599220383175,
     "sustainable": true,
-    "buffer_seconds": 60.71074079350998,
+    "buffer_seconds": 60.37008630239775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387154.30077657,
-    "upload_bytes_per_second": 5617326.315589492,
+    "generation_bytes_per_second": 403659.98668166425,
+    "upload_bytes_per_second": 5578018.702668729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.07430045578789,
+    "estimated_full_shards_per_hour": 86.61603641831823,
     "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": 1.270394237999426,
-      "bytes": 10514630,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.509260009999707,
       "bytes": 6276691,
@@ -562,6 +556,12 @@
       "bytes": 1165763,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.215623644999141,
+      "bytes": 7281440,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 334100, "loss": 0.6207200642675161, "lr": 0.0003, "grad_norm": 0.18417173892827268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48903975511, "tokens_per_second": 147978.44412231215, "data_wait_s": 1.3945998944109306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334200, "loss": 0.583866011351347, "lr": 0.0003, "grad_norm": 0.1469988414683677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48918607559, "tokens_per_second": 147891.47776618035, "data_wait_s": 1.4023997209733352e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334300, "loss": 0.6059367917478085, "lr": 0.0003, "grad_norm": 0.16072774066710382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48933096580, "tokens_per_second": 147556.20351281943, "data_wait_s": 1.0681997082429007e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 334400, "loss": 0.5995811820030212, "lr": 0.0003, "grad_norm": 0.18325775527286578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48947710891, "tokens_per_second": 147655.47456030236, "data_wait_s": 1.2196000170661137e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6525c287744c7a6e4b7cad64c9fd3d2031cd1f9efec3e0bf084b707619b281a5
-size 411288
+oid sha256:1e16000d5df3307782998d5b8d967d5d498856704b8f5d3502fbff57c519074e
+size 413406