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": 75718.46387726085,
-      "required_pages_per_second_upper_estimate": 0.006563084542188769
+      "required_token_positions_per_second": 75476.08801200095,
+      "required_pages_per_second_upper_estimate": 0.006542076016484056
     },
     {
       "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": 9464.807984657606,
-      "required_pages_per_second_upper_estimate": 0.003775653416569972
+      "required_token_positions_per_second": 9434.51100150012,
+      "required_pages_per_second_upper_estimate": 0.0037635674970081853
     },
     {
       "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": 18929.61596931521,
-      "required_pages_per_second_upper_estimate": 0.004379033504638495
+      "required_token_positions_per_second": 18869.02200300024,
+      "required_pages_per_second_upper_estimate": 0.004365016156948911
     },
     {
       "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": 28394.42395397282,
-      "required_pages_per_second_upper_estimate": 0.0017290714447677777
+      "required_token_positions_per_second": 28303.533004500354,
+      "required_pages_per_second_upper_estimate": 0.0017235366628128622
     },
     {
       "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": 28394.42395397282,
-      "required_pages_per_second_upper_estimate": 0.002185881356773007
+      "required_token_positions_per_second": 28303.533004500354,
+      "required_pages_per_second_upper_estimate": 0.002178884319880365
     },
     {
       "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": 28394.42395397282,
-      "required_pages_per_second_upper_estimate": 0.35256371548447074
+      "required_token_positions_per_second": 28303.533004500354,
+      "required_pages_per_second_upper_estimate": 0.35143515408446246
     }
   ],
   "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": 0.1215175708903189,
-    "producer_p95_seconds": 0.39656834000197705,
-    "gpu_update_seconds": 0.9737123050926207,
-    "producer_batches_per_second": 8.229262588721385,
-    "gpu_batches_per_second": 1.026997394168577,
-    "producer_headroom": 8.012934244476366,
+    "producer_batch_seconds": 0.10216759300766398,
+    "producer_p95_seconds": 0.3933907550017466,
+    "gpu_update_seconds": 0.9768391810168673,
+    "producer_batches_per_second": 9.787839475918613,
+    "gpu_batches_per_second": 1.0237099611002733,
+    "producer_headroom": 9.561145097580901,
     "sustainable": true,
-    "buffer_seconds": 60.37016291574248,
+    "buffer_seconds": 60.56402922304577,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341217.9919723042,
-    "upload_bytes_per_second": 3824891.637189163,
+    "generation_bytes_per_second": 372501.2499438568,
+    "upload_bytes_per_second": 3871341.464317895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.21743792893261,
+    "estimated_full_shards_per_hour": 79.93009685265329,
     "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": 0.9436809830003767,
-      "bytes": 889392,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1113080279974383,
       "bytes": 11835067,
@@ -562,6 +556,12 @@
       "bytes": 12565360,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.941881225000543,
+      "bytes": 6023516,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 35300, "loss": 0.6423078514635563, "lr": 0.0003, "grad_norm": 0.13669549508507228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5165289678, "tokens_per_second": 145017.1141338006, "data_wait_s": 1.1691998224705458e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35400, "loss": 0.6142779737710953, "lr": 0.0003, "grad_norm": 0.13293723371427815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5180032624, "tokens_per_second": 149743.53608937512, "data_wait_s": 1.138600055128336e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35500, "loss": 0.6256066113710403, "lr": 0.0003, "grad_norm": 0.1355154730981297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5194677629, "tokens_per_second": 149382.83431856887, "data_wait_s": 9.512004908174276e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 35600, "loss": 0.5915225241333246, "lr": 0.0003, "grad_norm": 0.1300828179687159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5209305680, "tokens_per_second": 148522.84898063625, "data_wait_s": 1.2244003301020712e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c33538bff5aee0a3b132be10b653e3c8da4ff5ffcc6a584f79962cb328138b51
-size 751762
+oid sha256:2b3c6e45de0c8e72ef4698c0962ea588fc6cf4510657b946700764d478d9c3b3
+size 753880