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": 75791.10030056722,
-      "required_pages_per_second_upper_estimate": 0.006569380483265635
+      "required_token_positions_per_second": 75189.84615559765,
+      "required_pages_per_second_upper_estimate": 0.006517265297844371
     },
     {
       "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": 9473.887537570903,
-      "required_pages_per_second_upper_estimate": 0.0037792753859784995
+      "required_token_positions_per_second": 9398.730769449707,
+      "required_pages_per_second_upper_estimate": 0.003749294227481134
     },
     {
       "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": 18947.775075141806,
-      "required_pages_per_second_upper_estimate": 0.004383234294182131
+      "required_token_positions_per_second": 18797.461538899413,
+      "required_pages_per_second_upper_estimate": 0.004348461903000338
     },
     {
       "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": 28421.66261271271,
-      "required_pages_per_second_upper_estimate": 0.0017307301361748394
+      "required_token_positions_per_second": 28196.19230834912,
+      "required_pages_per_second_upper_estimate": 0.0017170001775903606
     },
     {
       "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": 28421.66261271271,
-      "required_pages_per_second_upper_estimate": 0.002187978264124237
+      "required_token_positions_per_second": 28196.19230834912,
+      "required_pages_per_second_upper_estimate": 0.0021706209359525787
     },
     {
       "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": 28421.66261271271,
-      "required_pages_per_second_upper_estimate": 0.35290192846409363
+      "required_token_positions_per_second": 28196.19230834912,
+      "required_pages_per_second_upper_estimate": 0.35010234188446454
     }
   ],
   "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": 0.2305048603591331,
-    "producer_p95_seconds": 0.22967742699984228,
-    "gpu_update_seconds": 0.9727791219234775,
-    "producer_batches_per_second": 4.3383033157824595,
-    "gpu_batches_per_second": 1.0279825887121206,
-    "producer_headroom": 4.220210890164572,
+    "producer_batch_seconds": 0.1265878362577837,
+    "producer_p95_seconds": 0.3200532730006671,
+    "gpu_update_seconds": 0.9805579312853957,
+    "producer_batches_per_second": 7.899653154380475,
+    "gpu_batches_per_second": 1.01982755744897,
+    "producer_headroom": 7.746067554931469,
     "sustainable": true,
-    "buffer_seconds": 60.312305559255606,
+    "buffer_seconds": 60.79459173969453,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372042.93856370647,
-    "upload_bytes_per_second": 5846084.839141798,
+    "generation_bytes_per_second": 323902.76741566544,
+    "upload_bytes_per_second": 5847071.590211856,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.83175389941593,
+    "estimated_full_shards_per_hour": 69.50199381687615,
     "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": 1.1859958829991228,
-      "bytes": 7375397,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 0.9071520359993883,
       "bytes": 6539230,
@@ -562,6 +556,12 @@
       "bytes": 10705946,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0223479320011393,
+      "bytes": 6489223,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"step": 332400, "loss": 0.5818393640220165, "lr": 0.0003, "grad_norm": 0.14548106083418996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48654888422, "tokens_per_second": 150171.41094323076, "data_wait_s": 1.72869986272417e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332500, "loss": 0.6297379173338413, "lr": 0.0003, "grad_norm": 0.19845924071430546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48669734940, "tokens_per_second": 151188.29437379967, "data_wait_s": 1.0543000826146454e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332600, "loss": 0.582495553418994, "lr": 0.0003, "grad_norm": 0.14115508804723378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48684481372, "tokens_per_second": 129070.67810782626, "data_wait_s": 1.3593999028671533e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 332700, "loss": 0.6097974088042974, "lr": 0.0003, "grad_norm": 0.1485296390404382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48699049698, "tokens_per_second": 148018.8358762755, "data_wait_s": 1.3437998859444633e-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 sha256:932cc2bb76efab4cd773d04dcd7b15a5baf7eed3a0e0d9d9fd6e4a584c5e85c5
-size 375282
+oid sha256:f255b773dfbe8210e38f0c7916c9671f3eb00650a8b19353a21af92db1889c51
+size 377400