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": 76068.09712225226,
-      "required_pages_per_second_upper_estimate": 0.0065933898657272675
+      "required_token_positions_per_second": 75858.13281100985,
+      "required_pages_per_second_upper_estimate": 0.006575190691378457
     },
     {
       "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": 9508.512140281533,
-      "required_pages_per_second_upper_estimate": 0.0037930876576837136
+      "required_token_positions_per_second": 9482.266601376232,
+      "required_pages_per_second_upper_estimate": 0.0037826179198086132
     },
     {
       "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": 19017.024280563066,
-      "required_pages_per_second_upper_estimate": 0.004399253879111949
+      "required_token_positions_per_second": 18964.533202752464,
+      "required_pages_per_second_upper_estimate": 0.0043871109920718835
     },
     {
       "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": 28525.5364208446,
-      "required_pages_per_second_upper_estimate": 0.001737055506106319
+      "required_token_positions_per_second": 28446.799804128696,
+      "required_pages_per_second_upper_estimate": 0.0017322608592474221
     },
     {
       "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": 28525.5364208446,
-      "required_pages_per_second_upper_estimate": 0.0021959747574153343
+      "required_token_positions_per_second": 28446.799804128696,
+      "required_pages_per_second_upper_estimate": 0.002189913394703639
     },
     {
       "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": 28525.5364208446,
-      "required_pages_per_second_upper_estimate": 0.35419169351781915
+      "required_token_positions_per_second": 28446.799804128696,
+      "required_pages_per_second_upper_estimate": 0.35321404825271235
     }
   ],
   "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": 0.0896060322578478,
-    "producer_p95_seconds": 0.25881659500009846,
-    "gpu_update_seconds": 0.9692368126615367,
-    "producer_batches_per_second": 11.15996294895,
-    "gpu_batches_per_second": 1.0317395985548539,
-    "producer_headroom": 10.816646918061142,
+    "producer_batch_seconds": 0.5191029269453225,
+    "producer_p95_seconds": 0.3569480099977227,
+    "gpu_update_seconds": 0.971919519607518,
+    "producer_batches_per_second": 1.926400234120296,
+    "gpu_batches_per_second": 1.0288917753229418,
+    "producer_headroom": 1.8723059901180081,
     "sustainable": true,
-    "buffer_seconds": 60.09268238501528,
+    "buffer_seconds": 60.259010215666116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450484.42988051823,
-    "upload_bytes_per_second": 5444833.188546474,
+    "generation_bytes_per_second": 374753.9204028042,
+    "upload_bytes_per_second": 5572694.160781924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.66347191154156,
+    "estimated_full_shards_per_hour": 80.41346749365897,
     "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": 2.058075558999917,
-      "bytes": 7841052,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1063804570003413,
       "bytes": 836703,
@@ -562,6 +556,12 @@
       "bytes": 5895393,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2639917810010957,
+      "bytes": 13243059,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 177000, "loss": 0.6187183447182178, "lr": 0.0003, "grad_norm": 0.13996301249884616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25904777042, "tokens_per_second": 149623.20916157932, "data_wait_s": 9.587001841282472e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 177100, "loss": 0.6259839870035648, "lr": 0.0003, "grad_norm": 0.14912587300424573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25919474028, "tokens_per_second": 150611.38927202567, "data_wait_s": 1.5474997780984268e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177200, "loss": 0.5881300400942564, "lr": 0.0003, "grad_norm": 0.1371592981511136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25934216054, "tokens_per_second": 138680.48862020616, "data_wait_s": 1.2162003258708864e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 177300, "loss": 0.5884219035506248, "lr": 0.0003, "grad_norm": 0.13482742427755728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25948848126, "tokens_per_second": 149604.35016829346, "data_wait_s": 1.1733998690033332e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fcee5dc81c28baa52f821df6a825226be5cc42129e256be8d1680e3c3c4fc26
-size 473493
+oid sha256:e81205539b36f430c4a059161393e853c8a857a2d8183cd1df45ea639a9ed98e
+size 475611