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": 75585.75200563263,
-      "required_pages_per_second_upper_estimate": 0.006551581413511199
+      "required_token_positions_per_second": 75622.80258898545,
+      "required_pages_per_second_upper_estimate": 0.006554792864172376
     },
     {
       "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": 9448.219000704079,
-      "required_pages_per_second_upper_estimate": 0.0037690358228434973
+      "required_token_positions_per_second": 9452.85032362318,
+      "required_pages_per_second_upper_estimate": 0.0037708833268003754
     },
     {
       "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": 18896.438001408158,
-      "required_pages_per_second_upper_estimate": 0.00437135836567549
+      "required_token_positions_per_second": 18905.70064724636,
+      "required_pages_per_second_upper_estimate": 0.004373501115773159
     },
     {
       "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": 28344.65700211224,
-      "required_pages_per_second_upper_estimate": 0.0017260408985064854
+      "required_token_positions_per_second": 28358.550970869543,
+      "required_pages_per_second_upper_estimate": 0.0017268869683077848
     },
     {
       "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": 28344.65700211224,
-      "required_pages_per_second_upper_estimate": 0.0021820501590550396
+      "required_token_positions_per_second": 28358.550970869543,
+      "required_pages_per_second_upper_estimate": 0.002183119755230944
     },
     {
       "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": 28344.65700211224,
-      "required_pages_per_second_upper_estimate": 0.3519457765016358
+      "required_token_positions_per_second": 28358.550970869543,
+      "required_pages_per_second_upper_estimate": 0.3521182930934793
     }
   ],
   "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": 0.059411653953219457,
-    "producer_p95_seconds": 0.2047351860001072,
-    "gpu_update_seconds": 0.9754219286527149,
-    "producer_batches_per_second": 16.831714545220315,
-    "gpu_batches_per_second": 1.025197374208342,
-    "producer_headroom": 16.418023464230757,
+    "producer_batch_seconds": 0.11254851542173583,
+    "producer_p95_seconds": 0.25166185700072674,
+    "gpu_update_seconds": 0.9749440311107773,
+    "producer_batches_per_second": 8.885057224014488,
+    "gpu_batches_per_second": 1.0256999049070292,
+    "producer_headroom": 8.662433506630618,
     "sustainable": true,
-    "buffer_seconds": 60.476159576468326,
+    "buffer_seconds": 60.446529928868195,
     "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": 442893.7858211233,
-    "upload_bytes_per_second": 5968922.143034727,
+    "generation_bytes_per_second": 441099.0674377675,
+    "upload_bytes_per_second": 5981354.874411023,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.03469639754556,
+    "estimated_full_shards_per_hour": 94.64959161138314,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 5949961,
       "pages": 40,
       "shards": 2
+    },
+    {
+      "seconds": 0.8509849889996985,
+      "bytes": 5837001,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 126300, "loss": 0.5899433512240648, "lr": 0.0003, "grad_norm": 0.1420256776911786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18483034000, "tokens_per_second": 147523.62749179584, "data_wait_s": 1.979500120796729e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126400, "loss": 0.5922726560384035, "lr": 0.0003, "grad_norm": 0.15503752254612066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18497771948, "tokens_per_second": 150291.41941792518, "data_wait_s": 1.3797000065096654e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126500, "loss": 0.5903206244111061, "lr": 0.0003, "grad_norm": 0.1325108319875284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18512600986, "tokens_per_second": 151029.20525673678, "data_wait_s": 1.2004999007331207e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 126600, "loss": 0.6114125605672598, "lr": 0.0003, "grad_norm": 0.1337470624611966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18527324149, "tokens_per_second": 149839.31787375326, "data_wait_s": 1.622099989617709e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2988359d63e89c4fc9b48e96d95ab2cb469fedc836d7972c478fadcc0a54329
-size 245301
+oid sha256:4c43c8f4bc606694779c30376941cdda91a3457625feb4137546ef6ffc27906f
+size 247419