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": 74158.06268287537,
-      "required_pages_per_second_upper_estimate": 0.0064278329214601075
+      "required_token_positions_per_second": 74000.45746902973,
+      "required_pages_per_second_upper_estimate": 0.006414172100970729
     },
     {
       "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": 9269.757835359422,
-      "required_pages_per_second_upper_estimate": 0.003697844995755314
+      "required_token_positions_per_second": 9250.057183628716,
+      "required_pages_per_second_upper_estimate": 0.0036899861112289437
     },
     {
       "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": 18539.515670718843,
-      "required_pages_per_second_upper_estimate": 0.004288790666088994
+      "required_token_positions_per_second": 18500.114367257433,
+      "required_pages_per_second_upper_estimate": 0.004279675867972453
     },
     {
       "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": 27809.27350607827,
-      "required_pages_per_second_upper_estimate": 0.001693438852538134
+      "required_token_positions_per_second": 27750.17155088615,
+      "required_pages_per_second_upper_estimate": 0.0016898398535509283
     },
     {
       "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": 27809.27350607827,
-      "required_pages_per_second_upper_estimate": 0.002140834855493973
+      "required_token_positions_per_second": 27750.17155088615,
+      "required_pages_per_second_upper_estimate": 0.0021362850233786
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27809.27350607827,
-      "required_pages_per_second_upper_estimate": 0.3452981052941911
+      "required_token_positions_per_second": 27750.17155088615,
+      "required_pages_per_second_upper_estimate": 0.3445642568122248
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10907335314841049,
-    "producer_p95_seconds": 0.2211360560004323,
-    "gpu_update_seconds": 0.9942007292623801,
-    "producer_batches_per_second": 9.168142090940869,
-    "gpu_batches_per_second": 1.0058330984547985,
-    "producer_headroom": 9.114973552794535,
+    "producer_batch_seconds": 0.07868700189060007,
+    "producer_p95_seconds": 0.2264933309998014,
+    "gpu_update_seconds": 0.9963181650715638,
+    "producer_batches_per_second": 12.70857925671533,
+    "gpu_batches_per_second": 1.0036954409319354,
+    "producer_headroom": 12.661788365717156,
     "sustainable": true,
-    "buffer_seconds": 60.64624448500519,
+    "buffer_seconds": 60.77540806936539,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 74,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286874.92624424014,
-    "upload_bytes_per_second": 5267017.221117124,
+    "generation_bytes_per_second": 296653.99699928664,
+    "upload_bytes_per_second": 5321500.534583574,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.55668106551555,
+    "estimated_full_shards_per_hour": 63.6550420044322,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 5899615,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9532060609999462,
+      "bytes": 7422959,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -70,3 +70,4 @@
 {"step": 202000, "loss": 0.5699166264384985, "lr": 0.0003, "grad_norm": 0.14428473388935445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29564836318, "tokens_per_second": 146408.9944625575, "data_wait_s": 1.1482999980216846e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202100, "loss": 0.6176916379481554, "lr": 0.0003, "grad_norm": 0.16096172026217562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29579465932, "tokens_per_second": 148151.25711797716, "data_wait_s": 1.4909000128682237e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202200, "loss": 0.5846894299611449, "lr": 0.0003, "grad_norm": 0.14067985833372987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29594124743, "tokens_per_second": 148035.92941285914, "data_wait_s": 2.010000025620684e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 202300, "loss": 0.6223247908055782, "lr": 0.0003, "grad_norm": 0.1508217402394871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29608864297, "tokens_per_second": 148892.05627461962, "data_wait_s": 2.1283999558363575e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e88ab3b7d824fcb25110e4de4f2e2a9b91acc9eb6dd84660354dcbe3f7378cac
-size 153444
+oid sha256:f95a789f6d5eb3f30541ce8fe1c053a8ed2f66ff903e90c385f5e3fd34257363
+size 155562