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": 75594.23205540983,
-      "required_pages_per_second_upper_estimate": 0.006552316442733397
+      "required_token_positions_per_second": 75389.17737919465,
+      "required_pages_per_second_upper_estimate": 0.006534542823105391
     },
     {
       "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": 9449.279006926228,
-      "required_pages_per_second_upper_estimate": 0.003769458675174018
+      "required_token_positions_per_second": 9423.64717239933,
+      "required_pages_per_second_upper_estimate": 0.003759233753151161
     },
     {
       "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": 18898.558013852456,
-      "required_pages_per_second_upper_estimate": 0.004371848793243534
+      "required_token_positions_per_second": 18847.29434479866,
+      "required_pages_per_second_upper_estimate": 0.0043599898456177
     },
     {
       "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": 28347.837020778687,
-      "required_pages_per_second_upper_estimate": 0.0017262345449590036
+      "required_token_positions_per_second": 28270.941517197993,
+      "required_pages_per_second_upper_estimate": 0.0017215520122304672
     },
     {
       "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": 28347.837020778687,
-      "required_pages_per_second_upper_estimate": 0.002182294965694838
+      "required_token_positions_per_second": 28270.941517197993,
+      "required_pages_per_second_upper_estimate": 0.0021763753369688182
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28347.837020778687,
-      "required_pages_per_second_upper_estimate": 0.3519852616906352
+      "required_token_positions_per_second": 28270.941517197993,
+      "required_pages_per_second_upper_estimate": 0.35103047688885847
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.4394517487734788,
-    "producer_p95_seconds": 0.29664762699758285,
-    "gpu_update_seconds": 0.9753125072552903,
-    "producer_batches_per_second": 2.275562681889481,
-    "gpu_batches_per_second": 1.0253123922445995,
-    "producer_headroom": 2.2193847446902026,
+    "producer_batch_seconds": 0.09659543163297712,
+    "producer_p95_seconds": 0.2805416939991119,
+    "gpu_update_seconds": 0.9779653069983877,
+    "producer_batches_per_second": 10.352456457771092,
+    "gpu_batches_per_second": 1.0225311601995801,
+    "producer_headroom": 10.124343257911546,
     "sustainable": true,
-    "buffer_seconds": 60.469375449828,
+    "buffer_seconds": 60.633849033900034,
     "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": 291347.2305935843,
-    "upload_bytes_per_second": 5190797.977280912,
+    "generation_bytes_per_second": 291627.86195221404,
+    "upload_bytes_per_second": 5204555.524792469,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.516333468967886,
+    "estimated_full_shards_per_hour": 62.57655042576614,
     "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": 1.1652991789997031,
-      "bytes": 7620124,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0473349340009008,
       "bytes": 5879474,
@@ -562,6 +556,12 @@
       "bytes": 762678,
       "pages": 21,
       "shards": 1
+    },
+    {
+      "seconds": 1.7635130470007425,
+      "bytes": 11843733,
+      "pages": 46,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 378000, "loss": 0.6229256056249142, "lr": 0.0003, "grad_norm": 0.18372337081710355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55328384108, "tokens_per_second": 148847.05696519098, "data_wait_s": 1.2398999388096854e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378100, "loss": 0.5834005698561668, "lr": 0.0003, "grad_norm": 0.1435049871246068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55343008177, "tokens_per_second": 146524.02061653914, "data_wait_s": 1.0071998985949904e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378200, "loss": 0.5748591497540474, "lr": 0.0003, "grad_norm": 0.15164900917258697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55357767242, "tokens_per_second": 149328.52792212137, "data_wait_s": 1.9019000319531187e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 378300, "loss": 0.557722071185708, "lr": 0.0003, "grad_norm": 0.163479854615462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55372456151, "tokens_per_second": 150226.34467800736, "data_wait_s": 1.0760002624010667e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:438b9dd658c4b651639fdb6a233c9f3d5c94bbf59ffb7e047b646b3924e20e36
-size 494673
+oid sha256:481b0294290399f9d50d3aebd2bbf4bb54ac45bb921c0f05cbe5e72d01ba8ba9
+size 496791