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": 76925.05901582258,
-      "required_pages_per_second_upper_estimate": 0.005321854073442747
+      "required_token_positions_per_second": 76849.74134771495,
+      "required_pages_per_second_upper_estimate": 0.005316643422401993
     },
     {
       "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": 9615.632376977823,
-      "required_pages_per_second_upper_estimate": 0.004025864479003168
+      "required_token_positions_per_second": 9606.21766846437,
+      "required_pages_per_second_upper_estimate": 0.004021922737150055
     },
     {
       "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": 19231.264753955646,
-      "required_pages_per_second_upper_estimate": 0.004005365285692194
+      "required_token_positions_per_second": 19212.43533692874,
+      "required_pages_per_second_upper_estimate": 0.0040014436146906095
     },
     {
       "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": 28846.89713093347,
-      "required_pages_per_second_upper_estimate": 0.001729327436337927
+      "required_token_positions_per_second": 28818.65300539311,
+      "required_pages_per_second_upper_estimate": 0.001727634244138065
     },
     {
       "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": 28846.89713093347,
-      "required_pages_per_second_upper_estimate": 0.0022427193678391222
+      "required_token_positions_per_second": 28818.65300539311,
+      "required_pages_per_second_upper_estimate": 0.0022405235113111383
     },
     {
       "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": 28846.89713093347,
-      "required_pages_per_second_upper_estimate": 0.37001702301065237
+      "required_token_positions_per_second": 28818.65300539311,
+      "required_pages_per_second_upper_estimate": 0.3696547376943999
     }
   ],
   "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": 0.10532351478926216,
-    "producer_p95_seconds": 0.25544874299885123,
-    "gpu_update_seconds": 0.9584393036973168,
-    "producer_batches_per_second": 9.494555911857502,
-    "gpu_batches_per_second": 1.043362888126934,
-    "producer_headroom": 9.099955557075946,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.381676132930956,
+    "buffer_seconds": 60.44085404248546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 95379.97057390731,
-    "upload_bytes_per_second": 4986623.067488692,
+    "generation_bytes_per_second": 101578.17445152043,
+    "upload_bytes_per_second": 5002432.148934043,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 20.46632135308184,
+    "estimated_full_shards_per_hour": 21.796311618415924,
     "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": 1.7537132789984753,
-      "bytes": 11388396,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.1776387320023787,
       "bytes": 5920541,
@@ -562,6 +556,12 @@
       "bytes": 863102,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.423836862002645,
+      "bytes": 10998784,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -268,3 +268,4 @@
 {"step": 711800, "loss": 0.5453563332557678, "lr": 8e-05, "grad_norm": 0.2102621463966983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104187343352, "tokens_per_second": 150597.08948802997, "data_wait_s": 1.7242000467376783e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:9b8c9814a7868eef6a0836d60f16b8563a41270fa4b4e5c12b8de4349314ca24
-size 575940
+oid sha256:ba8b54d1b6d1f86cbfcea889cbd292c78e65b6c821d42d2b0edde7163f37bce3
+size 578058