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": 75355.17591607226,
-      "required_pages_per_second_upper_estimate": 0.006531595662457862
+      "required_token_positions_per_second": 75896.92466537176,
+      "required_pages_per_second_upper_estimate": 0.006578553070997491
     },
     {
       "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": 9419.396989509032,
-      "required_pages_per_second_upper_estimate": 0.003757538291650324
+      "required_token_positions_per_second": 9487.11558317147,
+      "required_pages_per_second_upper_estimate": 0.003784552251145472
     },
     {
       "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": 18838.793979018064,
-      "required_pages_per_second_upper_estimate": 0.004358023435595686
+      "required_token_positions_per_second": 18974.23116634294,
+      "required_pages_per_second_upper_estimate": 0.00438935444527021
     },
     {
       "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": 28258.1909685271,
-      "required_pages_per_second_upper_estimate": 0.0017207755707133686
+      "required_token_positions_per_second": 28461.34674951441,
+      "required_pages_per_second_upper_estimate": 0.0017331466918994883
     },
     {
       "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": 28258.1909685271,
-      "required_pages_per_second_upper_estimate": 0.0021753937644363552
+      "required_token_positions_per_second": 28461.34674951441,
+      "required_pages_per_second_upper_estimate": 0.0021910332588279536
     },
     {
       "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": 28258.1909685271,
-      "required_pages_per_second_upper_estimate": 0.35087215774770725
+      "required_token_positions_per_second": 28461.34674951441,
+      "required_pages_per_second_upper_estimate": 0.35339467262890856
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07829866999992419,
-    "producer_p95_seconds": 0.28997837000133586,
-    "gpu_update_seconds": 0.9784065806191661,
-    "producer_batches_per_second": 12.77160901968026,
-    "gpu_batches_per_second": 1.022069985840824,
-    "producer_headroom": 12.495826309950264,
+    "producer_batch_seconds": 0.24029638742965176,
+    "producer_p95_seconds": 0.3282314349999069,
+    "gpu_update_seconds": 0.9714227595527156,
+    "producer_batches_per_second": 4.161527398295807,
+    "gpu_batches_per_second": 1.0294179235212098,
+    "producer_headroom": 4.042602429206746,
     "sustainable": true,
-    "buffer_seconds": 60.6612079983883,
+    "buffer_seconds": 60.22821109226837,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519624.42624130246,
-    "upload_bytes_per_second": 4083111.684588883,
+    "generation_bytes_per_second": 511750.5907796016,
+    "upload_bytes_per_second": 4025015.123515209,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.49930563382439,
+    "estimated_full_shards_per_hour": 109.809763837252,
     "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": 1.1724209370004246,
-      "bytes": 11575534,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.0102253660006681,
       "bytes": 11368590,
@@ -562,6 +556,12 @@
       "bytes": 7368748,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1227261329986504,
+      "bytes": 5278941,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -210,3 +210,4 @@
 {"step": 136000, "loss": 0.6349450200796127, "lr": 0.0003, "grad_norm": 0.14239278688334917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19903261936, "tokens_per_second": 150953.43991890686, "data_wait_s": 9.090999810723588e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 136100, "loss": 0.6483194530010223, "lr": 0.0003, "grad_norm": 0.16793504180325042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19917980330, "tokens_per_second": 149455.7620041273, "data_wait_s": 1.1936001101275906e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136200, "loss": 0.6191393472254276, "lr": 0.0003, "grad_norm": 0.13696342735562392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19932615125, "tokens_per_second": 149543.99051949452, "data_wait_s": 1.565600177855231e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 136300, "loss": 0.6153386607766151, "lr": 0.0003, "grad_norm": 0.12918039986370972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19947345253, "tokens_per_second": 150388.36224613513, "data_wait_s": 9.712999599287286e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88d85a1b2775e10cb80ab522572c0bc0664c376ce0f30096a73b14b7cd8f5c3a
-size 452313
+oid sha256:efa468d01f0aa8553097af988d51704ecbd9f4377799b6d714d19b7ef411fbea
+size 454431