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": 76812.79570879469,
-      "required_pages_per_second_upper_estimate": 0.005314087437375796
+      "required_token_positions_per_second": 77070.0907769017,
+      "required_pages_per_second_upper_estimate": 0.005331887707194249
     },
     {
       "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": 9601.599463599336,
-      "required_pages_per_second_upper_estimate": 0.004019989191212149
+      "required_token_positions_per_second": 9633.761347112713,
+      "required_pages_per_second_upper_estimate": 0.004033454700222701
     },
     {
       "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": 19203.19892719867,
-      "required_pages_per_second_upper_estimate": 0.0039995199141243405
+      "required_token_positions_per_second": 19267.522694225427,
+      "required_pages_per_second_upper_estimate": 0.004012916858464204
     },
     {
       "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": 28804.798390798005,
-      "required_pages_per_second_upper_estimate": 0.0017268036811478602
+      "required_token_positions_per_second": 28901.284041338142,
+      "required_pages_per_second_upper_estimate": 0.0017325878485726833
     },
     {
       "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": 28804.798390798005,
-      "required_pages_per_second_upper_estimate": 0.0022394463759663782
+      "required_token_positions_per_second": 28901.284041338142,
+      "required_pages_per_second_upper_estimate": 0.0022469477109003486
     },
     {
       "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": 28804.798390798005,
-      "required_pages_per_second_upper_estimate": 0.3694770255743
+      "required_token_positions_per_second": 28901.284041338142,
+      "required_pages_per_second_upper_estimate": 0.37071463990120884
     }
   ],
   "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": 0.12028743826550681,
-    "producer_p95_seconds": 0.32645044200035045,
-    "gpu_update_seconds": 0.9598400802844169,
-    "producer_batches_per_second": 8.313420041357356,
-    "gpu_batches_per_second": 1.0418402195745806,
-    "producer_headroom": 7.979553759934525,
+    "producer_batch_seconds": 0.10892374962469376,
+    "producer_p95_seconds": 0.31112166800448904,
+    "gpu_update_seconds": 0.9566356968934134,
+    "producer_batches_per_second": 9.180734260852999,
+    "gpu_batches_per_second": 1.0453300072821954,
+    "producer_headroom": 8.782618117624345,
     "sustainable": true,
-    "buffer_seconds": 60.46992505791827,
+    "buffer_seconds": 60.268048904285045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424303.09910171427,
-    "upload_bytes_per_second": 4817714.282944879,
+    "generation_bytes_per_second": 424853.07215582626,
+    "upload_bytes_per_second": 4782926.039783627,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.04556779659816,
+    "estimated_full_shards_per_hour": 91.16357921129314,
     "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": 1.1343894960009493,
-      "bytes": 6177604,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 3.299507163999806,
       "bytes": 8285983,
@@ -562,6 +556,12 @@
       "bytes": 9866131,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0393312469968805,
+      "bytes": 2566382,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 794400, "loss": 0.5692605376243591, "lr": 3e-05, "grad_norm": 0.2803299069318174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116277915773, "tokens_per_second": 153034.29447523592, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 794500, "loss": 0.5542953014373779, "lr": 3e-05, "grad_norm": 0.2731092819199093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116292484074, "tokens_per_second": 151676.86563222043, "data_wait_s": 1.5102996258065104e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 794600, "loss": 0.508455753326416, "lr": 3e-05, "grad_norm": 0.2578542515966335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116307039693, "tokens_per_second": 151368.67870592087, "data_wait_s": 9.748000593390316e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 794700, "loss": 0.5018172860145569, "lr": 3e-05, "grad_norm": 0.27744146885200277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116321529952, "tokens_per_second": 151068.95150713107, "data_wait_s": 8.745002560317516e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86bfadad30d05946f9628f47692aa7640f1c685d200773ba35a3a939b447a731
-size 737691
+oid sha256:099ac8b0c37335b561c9c2d68ac5f3c7f088fca25844fac7a6fe4716e65c3cf6
+size 739809