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": 77294.54296758438,
-      "required_pages_per_second_upper_estimate": 0.005347415830546511
+      "required_token_positions_per_second": 77368.87838101329,
+      "required_pages_per_second_upper_estimate": 0.005352558526929446
     },
     {
       "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": 9661.817870948047,
-      "required_pages_per_second_upper_estimate": 0.004045201380865714
+      "required_token_positions_per_second": 9671.109797626661,
+      "required_pages_per_second_upper_estimate": 0.004049091716528556
     },
     {
       "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": 19323.635741896094,
-      "required_pages_per_second_upper_estimate": 0.004024603726493425
+      "required_token_positions_per_second": 19342.219595253322,
+      "required_pages_per_second_upper_estimate": 0.004028474253058565
     },
     {
       "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": 28985.453612844143,
-      "required_pages_per_second_upper_estimate": 0.001737633685864974
+      "required_token_positions_per_second": 29013.329392879983,
+      "required_pages_per_second_upper_estimate": 0.001739304796314271
     },
     {
       "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": 28985.453612844143,
-      "required_pages_per_second_upper_estimate": 0.0022534915248621215
+      "required_token_positions_per_second": 29013.329392879983,
+      "required_pages_per_second_upper_estimate": 0.00225565874414731
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28985.453612844143,
-      "required_pages_per_second_upper_estimate": 0.37179427679024313
+      "required_token_positions_per_second": 29013.329392879983,
+      "required_pages_per_second_upper_estimate": 0.3721518373658622
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14735919233595496,
-    "producer_p95_seconds": 0.2581016680014727,
-    "gpu_update_seconds": 0.9538577649772751,
-    "producer_batches_per_second": 6.786139257062178,
-    "gpu_batches_per_second": 1.0483743349553003,
-    "producer_headroom": 6.473011624565876,
+    "producer_batch_seconds": 0.08228636122652233,
+    "producer_p95_seconds": 0.2413801920010883,
+    "gpu_update_seconds": 0.9529413058945575,
+    "producer_batches_per_second": 12.15268223183604,
+    "gpu_batches_per_second": 1.0493825735271984,
+    "producer_headroom": 11.580792876127422,
     "sustainable": true,
-    "buffer_seconds": 60.09303919356834,
+    "buffer_seconds": 60.035302271357125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279313.02483269176,
-    "upload_bytes_per_second": 5008252.400212776,
+    "generation_bytes_per_second": 279028.5197498333,
+    "upload_bytes_per_second": 5042003.05856133,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.93407305465283,
+    "estimated_full_shards_per_hour": 59.87302488681077,
     "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": 1.0605444640004862,
-      "bytes": 2480385,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.4249819679989741,
       "bytes": 6270908,
@@ -562,6 +556,12 @@
       "bytes": 8233418,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1861695589977899,
+      "bytes": 6023568,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 904400, "loss": 0.5317763686180115, "lr": 3e-05, "grad_norm": 0.3131015411221213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132377999580, "tokens_per_second": 151253.96749319314, "data_wait_s": 1.4433000615099445e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904500, "loss": 0.5500661730766296, "lr": 3e-05, "grad_norm": 0.3273181986629234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132392803909, "tokens_per_second": 154384.530293512, "data_wait_s": 1.0823001503013074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904600, "loss": 0.5261452198028564, "lr": 3e-05, "grad_norm": 0.3217357915987881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132407413908, "tokens_per_second": 152818.99708768606, "data_wait_s": 1.5734000044176355e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 904700, "loss": 0.495917409658432, "lr": 3e-05, "grad_norm": 0.31905249510582967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132422138280, "tokens_per_second": 153116.43496323537, "data_wait_s": 1.2813001376343891e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afeeda1882f795b8a73a0a5819c5c68c39079e2f3d299ac57abfade09e7ff5c6
-size 417642
+oid sha256:e7c4fe93dcb2acc7f1d17fb481ab98dc2fc3ef0c2cfd3da57478b07732df04a1
+size 419760