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": 74789.22219739005,
-      "required_pages_per_second_upper_estimate": 0.0064825402285190745
+      "required_token_positions_per_second": 74454.43341493527,
+      "required_pages_per_second_upper_estimate": 0.0064535215853703054
     },
     {
       "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": 9348.652774673756,
-      "required_pages_per_second_upper_estimate": 0.0037293173666322625
+      "required_token_positions_per_second": 9306.804176866908,
+      "required_pages_per_second_upper_estimate": 0.0037126233352748157
     },
     {
       "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": 18697.30554934751,
-      "required_pages_per_second_upper_estimate": 0.004325292577502718
+      "required_token_positions_per_second": 18613.608353733816,
+      "required_pages_per_second_upper_estimate": 0.004305930704317823
     },
     {
       "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": 28045.958324021267,
-      "required_pages_per_second_upper_estimate": 0.0017078517161615934
+      "required_token_positions_per_second": 27920.412530600723,
+      "required_pages_per_second_upper_estimate": 0.0017002066360301542
     },
     {
       "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": 28045.958324021267,
-      "required_pages_per_second_upper_estimate": 0.0021590555079647345
+      "required_token_positions_per_second": 27920.412530600723,
+      "required_pages_per_second_upper_estimate": 0.0021493906452542217
     },
     {
       "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": 28045.958324021267,
-      "required_pages_per_second_upper_estimate": 0.3482369385999139
+      "required_token_positions_per_second": 27920.412530600723,
+      "required_pages_per_second_upper_estimate": 0.34667808002037226
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.42701781460147004,
+    "producer_batch_seconds": 0.13431586360158576,
     "producer_p95_seconds": 0.3386782650013629,
-    "gpu_update_seconds": 0.9858104929265185,
-    "producer_batches_per_second": 2.341822672979783,
-    "gpu_batches_per_second": 1.0143937472519267,
-    "producer_headroom": 2.308593363596697,
+    "gpu_update_seconds": 0.9902432483652539,
+    "producer_batches_per_second": 7.445136956914103,
+    "gpu_batches_per_second": 1.009852883774621,
+    "producer_headroom": 7.372496604738823,
     "sustainable": true,
-    "buffer_seconds": 60.13444006851763,
+    "buffer_seconds": 60.40483815028049,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297301.97635776654,
-    "upload_bytes_per_second": 5711591.078716606,
+    "generation_bytes_per_second": 295656.2448772032,
+    "upload_bytes_per_second": 5703058.435929407,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.79408329057453,
+    "estimated_full_shards_per_hour": 63.440947625513765,
     "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": 1.178703661000327,
-      "bytes": 7809750,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9307224249969295,
       "bytes": 6774165,
@@ -562,6 +556,12 @@
       "bytes": 5753379,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1451003039983334,
+      "bytes": 6976542,
+      "pages": 21,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 304800, "loss": 0.5849448200315237, "lr": 0.0003, "grad_norm": 0.16570698714419854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44614851586, "tokens_per_second": 143666.1496722523, "data_wait_s": 1.3000997569179162e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304900, "loss": 0.554118636995554, "lr": 0.0003, "grad_norm": 0.14330116719255986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44629402983, "tokens_per_second": 149150.65887491562, "data_wait_s": 1.2025000614812598e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 305000, "loss": 0.6317330747842789, "lr": 0.0003, "grad_norm": 0.15201925927561333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44643840707, "tokens_per_second": 147406.7625548003, "data_wait_s": 1.454499943065457e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 305100, "loss": 0.5901952311396599, "lr": 0.0003, "grad_norm": 0.1538424145560877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44658503019, "tokens_per_second": 126487.41882847535, "data_wait_s": 1.0969997674692422e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea9fbc6f279104c7a03c27b9ad3e37edea0a2d686b39ec28afdfd92ae04b5c3c
-size 529344
+oid sha256:440419eba2db3009f0b328654cd11d187ba90e05d7a7814cc47187ffa7139130
+size 531462