Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76169.05944255865,
-      "required_pages_per_second_upper_estimate": 0.005269552268803661
+      "required_token_positions_per_second": 75395.97940643643,
+      "required_pages_per_second_upper_estimate": 0.0052160687981119055
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9521.13243031983,
-      "required_pages_per_second_upper_estimate": 0.003986299324720754
+      "required_token_positions_per_second": 9424.497425804553,
+      "required_pages_per_second_upper_estimate": 0.0039458402662985725
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19042.26486063966,
-      "required_pages_per_second_upper_estimate": 0.003966001592176839
+      "required_token_positions_per_second": 18848.994851609106,
+      "required_pages_per_second_upper_estimate": 0.003925748546168662
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28563.397290959492,
-      "required_pages_per_second_upper_estimate": 0.0017123320538106781
+      "required_token_positions_per_second": 28273.49227741366,
+      "required_pages_per_second_upper_estimate": 0.0016949526909079305
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28563.397290959492,
-      "required_pages_per_second_upper_estimate": 0.0022206785022651564
+      "required_token_positions_per_second": 28273.49227741366,
+      "required_pages_per_second_upper_estimate": 0.00219813966261936
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28563.397290959492,
-      "required_pages_per_second_upper_estimate": 0.3663805914618783
+      "required_token_positions_per_second": 28273.49227741366,
+      "required_pages_per_second_upper_estimate": 0.362662001223864
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11040533194542945,
-    "producer_p95_seconds": 0.28076194499954,
-    "gpu_update_seconds": 0.9679520863140036,
-    "producer_batches_per_second": 9.057533566352344,
-    "gpu_batches_per_second": 1.0331089876649122,
-    "producer_headroom": 8.76725851240987,
+    "producer_batch_seconds": 0.4134950382734246,
+    "producer_p95_seconds": 0.28931141999782994,
+    "gpu_update_seconds": 0.9778770775369219,
+    "producer_batches_per_second": 2.4184087049159406,
+    "gpu_batches_per_second": 1.0226234185985843,
+    "producer_headroom": 2.364906436653052,
     "sustainable": true,
-    "buffer_seconds": 60.01302935146823,
+    "buffer_seconds": 60.62837880728916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320226.3162295755,
-    "upload_bytes_per_second": 5398746.312071206,
+    "generation_bytes_per_second": 292928.500611434,
+    "upload_bytes_per_second": 5416438.440388315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.7131129757447,
+    "estimated_full_shards_per_hour": 62.855637204716345,
     "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": 1.2023094430005585,
-      "bytes": 5878390,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9879657829987991,
       "bytes": 6822345,
@@ -562,6 +556,12 @@
       "bytes": 1155308,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.2314064579986734,
+      "bytes": 7429916,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 747300, "loss": 0.5461853742599487, "lr": 5e-05, "grad_norm": 0.23208668365497262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109384491561, "tokens_per_second": 148383.02233741563, "data_wait_s": 1.5613000869052485e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 747400, "loss": 0.5237144231796265, "lr": 5e-05, "grad_norm": 0.24144906531722096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109398952411, "tokens_per_second": 148961.94602863523, "data_wait_s": 1.5576999430777505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 747500, "loss": 0.5923118591308594, "lr": 5e-05, "grad_norm": 0.2593413670133543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109413518006, "tokens_per_second": 150019.92716357397, "data_wait_s": 9.978000889532268e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 747600, "loss": 0.49504661560058594, "lr": 5e-05, "grad_norm": 0.223179387904795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109428197268, "tokens_per_second": 149910.11415132481, "data_wait_s": 1.4328001270769164e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad3f31728634ccb9d41eab65150cf97368e4d1120e657c30fd3cf97249880392
-size 586530
+oid sha256:659c3d2d21482ec596e1b85e56c41f1bb5a4e8091f02adc194d390a2ab69300b
+size 588648