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": 77052.76774070505,
-      "required_pages_per_second_upper_estimate": 0.005330689259355183
+      "required_token_positions_per_second": 77288.03464984222,
+      "required_pages_per_second_upper_estimate": 0.005346965570023731
     },
     {
       "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": 9631.595967588131,
-      "required_pages_per_second_upper_estimate": 0.004032548101034025
+      "required_token_positions_per_second": 9661.004331230277,
+      "required_pages_per_second_upper_estimate": 0.004044860768774525
     },
     {
       "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": 19263.191935176263,
-      "required_pages_per_second_upper_estimate": 0.00401201487556406
+      "required_token_positions_per_second": 19322.008662460554,
+      "required_pages_per_second_upper_estimate": 0.004024264848755972
     },
     {
       "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": 28894.787902764394,
-      "required_pages_per_second_upper_estimate": 0.0017321984149842164
+      "required_token_positions_per_second": 28983.012993690827,
+      "required_pages_per_second_upper_estimate": 0.0017374873744733423
     },
     {
       "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": 28894.787902764394,
-      "required_pages_per_second_upper_estimate": 0.0022464426646997336
+      "required_token_positions_per_second": 28983.012993690827,
+      "required_pages_per_second_upper_estimate": 0.0022533017774581
     },
     {
       "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": 28894.787902764394,
-      "required_pages_per_second_upper_estimate": 0.37063131441059494
+      "required_token_positions_per_second": 28983.012993690827,
+      "required_pages_per_second_upper_estimate": 0.37176297114827705
     }
   ],
   "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": 0.11707211361719772,
-    "producer_p95_seconds": 0.208302150000236,
-    "gpu_update_seconds": 0.9568507681399139,
-    "producer_batches_per_second": 8.541743794511126,
-    "gpu_batches_per_second": 1.0450950485664205,
-    "producer_headroom": 8.173174111032314,
+    "producer_batch_seconds": 0.1843270247734381,
+    "producer_p95_seconds": 0.20115969700054848,
+    "gpu_update_seconds": 0.9539380880110208,
+    "producer_batches_per_second": 5.425140460163832,
+    "gpu_batches_per_second": 1.0482860602463406,
+    "producer_headroom": 5.1752481177599154,
     "sustainable": true,
-    "buffer_seconds": 60.28159839281457,
+    "buffer_seconds": 60.09809954469431,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459030.7204950226,
-    "upload_bytes_per_second": 3347614.017764525,
+    "generation_bytes_per_second": 469064.18514846254,
+    "upload_bytes_per_second": 3339894.55852635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.4973069299508,
+    "estimated_full_shards_per_hour": 100.6502548774758,
     "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": 1.2931400809993647,
-      "bytes": 6805061,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.581821284000398,
       "bytes": 6273764,
@@ -562,6 +556,12 @@
       "bytes": 2509117,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3743308060002164,
+      "bytes": 6082381,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 937200, "loss": 0.4811885356903076, "lr": 3e-05, "grad_norm": 0.3375159296900181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137179413409, "tokens_per_second": 151679.5245418816, "data_wait_s": 1.3246999515104108e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 937300, "loss": 0.5427599549293518, "lr": 3e-05, "grad_norm": 0.3539284067800814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137193896203, "tokens_per_second": 150954.3700951356, "data_wait_s": 1.467500078433659e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 937400, "loss": 0.5291251540184021, "lr": 3e-05, "grad_norm": 0.3317639549468237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137208485541, "tokens_per_second": 152619.0957799641, "data_wait_s": 1.417099883838091e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 937500, "loss": 0.5770094990730286, "lr": 3e-05, "grad_norm": 0.3401494716194435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137223151794, "tokens_per_second": 147682.17443085133, "data_wait_s": 1.4368999472935684e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ccb687e0b9c8473bb3e4eef03d7c6f201a4777d6e987d1445a8d4cda4650a1bf
-size 264363
+oid sha256:b229aff1a25425cdd3b6769bc162284ff5d0aa733281de81a4ca751be2c4f85d
+size 266481