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": 75609.98309263695,
-      "required_pages_per_second_upper_estimate": 0.006553681702719605
+      "required_token_positions_per_second": 75648.73732717025,
+      "required_pages_per_second_upper_estimate": 0.006557040821547243
     },
     {
       "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": 9451.247886579618,
-      "required_pages_per_second_upper_estimate": 0.0037702440907051295
+      "required_token_positions_per_second": 9456.092165896282,
+      "required_pages_per_second_upper_estimate": 0.0037721765461529765
     },
     {
       "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": 18902.495773159237,
-      "required_pages_per_second_upper_estimate": 0.004372759724556961
+      "required_token_positions_per_second": 18912.184331792563,
+      "required_pages_per_second_upper_estimate": 0.004375001002083973
     },
     {
       "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": 28353.743659738855,
-      "required_pages_per_second_upper_estimate": 0.0017265942282819382
+      "required_token_positions_per_second": 28368.276497688843,
+      "required_pages_per_second_upper_estimate": 0.0017274792018651846
     },
     {
       "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": 28353.743659738855,
-      "required_pages_per_second_upper_estimate": 0.0021827496751126146
+      "required_token_positions_per_second": 28368.276497688843,
+      "required_pages_per_second_upper_estimate": 0.002183868453207475
     },
     {
       "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": 28353.743659738855,
-      "required_pages_per_second_upper_estimate": 0.3520586023782715
+      "required_token_positions_per_second": 28368.276497688843,
+      "required_pages_per_second_upper_estimate": 0.3522390515873306
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.127990981242192,
-    "producer_p95_seconds": 0.39762537000206066,
-    "gpu_update_seconds": 0.9751093305981678,
-    "producer_batches_per_second": 7.813050500079703,
-    "gpu_batches_per_second": 1.0255260293597677,
-    "producer_headroom": 7.6185784430623995,
+    "producer_batch_seconds": 0.11795264328901567,
+    "producer_p95_seconds": 0.407388578998507,
+    "gpu_update_seconds": 0.9746097899973752,
+    "producer_batches_per_second": 8.477978721932761,
+    "gpu_batches_per_second": 1.0260516673064541,
+    "producer_headroom": 8.262721061785104,
     "sustainable": true,
-    "buffer_seconds": 60.456778497086404,
+    "buffer_seconds": 60.42580697983726,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399857.5445861867,
+    "generation_bytes_per_second": 458570.96408072504,
     "upload_bytes_per_second": 3802488.1757682385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.80012086095047,
+    "estimated_full_shards_per_hour": 98.39865390602411,
     "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": [

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

@@ -368,3 +368,4 @@
 {"step": 36800, "loss": 0.6500572897493839, "lr": 0.0003, "grad_norm": 0.12445203007592046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5385298056, "tokens_per_second": 151100.73728426103, "data_wait_s": 1.0128998837899417e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 36900, "loss": 0.6742070019245148, "lr": 0.0003, "grad_norm": 0.15172428596883727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5400003771, "tokens_per_second": 149008.34435460967, "data_wait_s": 1.195799995912239e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37000, "loss": 0.6387836299836636, "lr": 0.0003, "grad_norm": 0.12828591438246117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5414557540, "tokens_per_second": 147718.16022855372, "data_wait_s": 1.5706005797255784e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 37100, "loss": 0.6277224123477936, "lr": 0.0003, "grad_norm": 0.13178987176735782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5429011119, "tokens_per_second": 147283.71294513738, "data_wait_s": 1.1948999599553645e-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:eae2e1207ed97f304f02df68ddf81e9fd47aa35f5ee0545b1848b5fafb59945b
-size 783532
+oid sha256:82d9dc6c7454fceb0c81faf6eaa6265e3a93cd9150f89bf0fe2820a0940ff487
+size 785650