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": 76849.74134771495,
-      "required_pages_per_second_upper_estimate": 0.005316643422401993
+      "required_token_positions_per_second": 77070.01325026229,
+      "required_pages_per_second_upper_estimate": 0.005331882343721448
     },
     {
       "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": 9606.21766846437,
-      "required_pages_per_second_upper_estimate": 0.004021922737150055
+      "required_token_positions_per_second": 9633.751656282786,
+      "required_pages_per_second_upper_estimate": 0.004033450642874578
     },
     {
       "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": 19212.43533692874,
-      "required_pages_per_second_upper_estimate": 0.0040014436146906095
+      "required_token_positions_per_second": 19267.503312565572,
+      "required_pages_per_second_upper_estimate": 0.0040129128217755855
     },
     {
       "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": 28818.65300539311,
-      "required_pages_per_second_upper_estimate": 0.001727634244138065
+      "required_token_positions_per_second": 28901.254968848356,
+      "required_pages_per_second_upper_estimate": 0.0017325861057213118
     },
     {
       "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": 28818.65300539311,
-      "required_pages_per_second_upper_estimate": 0.0022405235113111383
+      "required_token_positions_per_second": 28901.254968848356,
+      "required_pages_per_second_upper_estimate": 0.002246945450642144
     },
     {
       "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": 28818.65300539311,
-      "required_pages_per_second_upper_estimate": 0.3696547376943999
+      "required_token_positions_per_second": 28901.254968848356,
+      "required_pages_per_second_upper_estimate": 0.37071426699052545
     }
   ],
   "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": 0.3755819342342477,
-    "producer_p95_seconds": 0.2799702560005244,
-    "gpu_update_seconds": 0.9593786355950074,
-    "producer_batches_per_second": 2.662534879476677,
-    "gpu_batches_per_second": 1.042341326873304,
-    "producer_headroom": 2.554379079896452,
+    "producer_batch_seconds": 0.13993730771878177,
+    "producer_p95_seconds": 0.2636863010011439,
+    "gpu_update_seconds": 0.9566366591970074,
+    "producer_batches_per_second": 7.146057161608408,
+    "gpu_batches_per_second": 1.0453289557598509,
+    "producer_headroom": 6.836180249511917,
     "sustainable": true,
-    "buffer_seconds": 60.44085404248546,
+    "buffer_seconds": 60.26810952941146,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 101578.17445152043,
-    "upload_bytes_per_second": 5002432.148934043,
+    "generation_bytes_per_second": 102459.62435056616,
+    "upload_bytes_per_second": 5020053.778750467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 21.796311618415924,
+    "estimated_full_shards_per_hour": 21.985450247647652,
     "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": 1.1776387320023787,
-      "bytes": 5920541,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.2547966029997042,
       "bytes": 8617085,
@@ -562,6 +556,12 @@
       "bytes": 10998784,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.2024595770017186,
+      "bytes": 7444430,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -269,3 +269,4 @@
 {"step": 711900, "loss": 0.5351439118385315, "lr": 8e-05, "grad_norm": 0.2392798639980104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104201837549, "tokens_per_second": 150747.37669953366, "data_wait_s": 2.061899795080535e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712000, "loss": 0.5082935094833374, "lr": 8e-05, "grad_norm": 0.208647967164889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104216434082, "tokens_per_second": 152052.86975688874, "data_wait_s": 1.3460998161463067e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712100, "loss": 0.5378831028938293, "lr": 8e-05, "grad_norm": 0.22318795780962136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104231235046, "tokens_per_second": 153819.13191602673, "data_wait_s": 1.5730998711660504e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 712200, "loss": 0.5908395648002625, "lr": 8e-05, "grad_norm": 0.22184218315466675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104245900717, "tokens_per_second": 152507.72395179849, "data_wait_s": 1.4692999684484676e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba8b54d1b6d1f86cbfcea889cbd292c78e65b6c821d42d2b0edde7163f37bce3
-size 578058
+oid sha256:c5c2d1bc616a882291447677f72775abea7b95ba36b86eaccb6b1f3b36c99471
+size 580176