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": 75281.53450616346,
-      "required_pages_per_second_upper_estimate": 0.006525212611689415
+      "required_token_positions_per_second": 75094.79425144794,
+      "required_pages_per_second_upper_estimate": 0.006509026439699522
     },
     {
       "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": 9410.191813270432,
-      "required_pages_per_second_upper_estimate": 0.003753866209219097
+      "required_token_positions_per_second": 9386.849281430992,
+      "required_pages_per_second_upper_estimate": 0.003744554524266392
     },
     {
       "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": 18820.383626540864,
-      "required_pages_per_second_upper_estimate": 0.004353764524561224
+      "required_token_positions_per_second": 18773.698562861984,
+      "required_pages_per_second_upper_estimate": 0.004342964756708173
     },
     {
       "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": 28230.5754398113,
-      "required_pages_per_second_upper_estimate": 0.001719093929902058
+      "required_token_positions_per_second": 28160.547844292974,
+      "required_pages_per_second_upper_estimate": 0.001714829616741391
     },
     {
       "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": 28230.5754398113,
-      "required_pages_per_second_upper_estimate": 0.0021732678445911366
+      "required_token_positions_per_second": 28160.547844292974,
+      "required_pages_per_second_upper_estimate": 0.002167876926439345
     },
     {
       "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": 28230.5754398113,
-      "required_pages_per_second_upper_estimate": 0.3505292653042862
+      "required_token_positions_per_second": 28160.547844292974,
+      "required_pages_per_second_upper_estimate": 0.34965975693523443
     }
   ],
   "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": 0.13080037027356184,
-    "producer_p95_seconds": 0.3714157790018362,
-    "gpu_update_seconds": 0.9793636711000322,
-    "producer_batches_per_second": 7.64523829640967,
-    "gpu_batches_per_second": 1.0210711602941007,
-    "producer_headroom": 7.487468644406331,
+    "producer_batch_seconds": 0.111315061632979,
+    "producer_p95_seconds": 0.368932520999806,
+    "gpu_update_seconds": 0.9817990812136544,
+    "producer_batches_per_second": 8.98351027551992,
+    "gpu_batches_per_second": 1.0185383334886058,
+    "producer_headroom": 8.820002134578882,
     "sustainable": true,
-    "buffer_seconds": 60.720547608202,
+    "buffer_seconds": 60.871543035246575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486187.7358656892,
-    "upload_bytes_per_second": 5997182.638333119,
+    "generation_bytes_per_second": 486404.0267885608,
+    "upload_bytes_per_second": 5991333.939715408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.32457024553305,
+    "estimated_full_shards_per_hour": 104.37098124258631,
     "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": 0.9731116549992294,
-      "bytes": 6198048,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2161200219998136,
       "bytes": 5847296,
@@ -562,6 +556,12 @@
       "bytes": 7625766,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.004631062001863,
+      "bytes": 5973803,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 114500, "loss": 0.6019764691591263, "lr": 0.0003, "grad_norm": 0.13809368823880785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16755176941, "tokens_per_second": 149165.51744536447, "data_wait_s": 2.0373001461848617e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114600, "loss": 0.5923062432557344, "lr": 0.0003, "grad_norm": 0.13503638414743294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16769917267, "tokens_per_second": 150322.0356558141, "data_wait_s": 1.4290999388322234e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114700, "loss": 0.6170772984623909, "lr": 0.0003, "grad_norm": 0.14149551904141514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16784645668, "tokens_per_second": 149495.65994409827, "data_wait_s": 1.5184996300376952e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 114800, "loss": 0.5815488835796714, "lr": 0.0003, "grad_norm": 0.1341264427182585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16799325296, "tokens_per_second": 149237.73573769786, "data_wait_s": 1.0627998562995344e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:695cddfda8cdd1e9e3780c2ac0e9fab14c4f8c3e23562012db28f52dfeaaddb8
-size 633126
+oid sha256:abbfad888d8183ae3be9e10add145cc42d83818a4391751e3579ccb41ca9bb7d
+size 635244