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": 75750.37559459335,
-      "required_pages_per_second_upper_estimate": 0.006565850568967643
+      "required_token_positions_per_second": 76148.47100009678,
+      "required_pages_per_second_upper_estimate": 0.006600356469753099
     },
     {
       "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": 9468.796949324169,
-      "required_pages_per_second_upper_estimate": 0.0037772446742158007
+      "required_token_positions_per_second": 9518.558875012097,
+      "required_pages_per_second_upper_estimate": 0.003797095450379806
     },
     {
       "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": 18937.593898648338,
-      "required_pages_per_second_upper_estimate": 0.004380879058182953
+      "required_token_positions_per_second": 19037.117750024194,
+      "required_pages_per_second_upper_estimate": 0.0044039021496387965
     },
     {
       "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": 28406.390847972503,
-      "required_pages_per_second_upper_estimate": 0.001729800165826918
+      "required_token_positions_per_second": 28555.67662503629,
+      "required_pages_per_second_upper_estimate": 0.0017388908864081625
     },
     {
       "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": 28406.390847972503,
-      "required_pages_per_second_upper_estimate": 0.0021868026013995854
+      "required_token_positions_per_second": 28555.67662503629,
+      "required_pages_per_second_upper_estimate": 0.0021982950337674357
     },
     {
       "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": 28406.390847972503,
-      "required_pages_per_second_upper_estimate": 0.35271230425732897
+      "required_token_positions_per_second": 28555.67662503629,
+      "required_pages_per_second_upper_estimate": 0.3545659339811054
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09593731933591698,
-    "producer_p95_seconds": 0.2937691020015336,
-    "gpu_update_seconds": 0.973302104725964,
-    "producer_batches_per_second": 10.423472397624314,
-    "gpu_batches_per_second": 1.0274302245360425,
-    "producer_headroom": 10.145187623160735,
+    "producer_batch_seconds": 0.0966948314842,
+    "producer_p95_seconds": 0.2635089439972944,
+    "gpu_update_seconds": 0.9682137938121739,
+    "producer_batches_per_second": 10.341814393289477,
+    "gpu_batches_per_second": 1.0328297390421113,
+    "producer_headroom": 10.013087348628149,
     "sustainable": true,
-    "buffer_seconds": 60.34473049300977,
+    "buffer_seconds": 60.02925521635478,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412242.5173852521,
+    "generation_bytes_per_second": 449012.3010405212,
     "upload_bytes_per_second": 5439793.333650127,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.45764771621869,
+    "estimated_full_shards_per_hour": 96.34758733188369,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 176600, "loss": 0.5876819044351578, "lr": 0.0003, "grad_norm": 0.1395021845340289, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25846084754, "tokens_per_second": 146814.96468439602, "data_wait_s": 1.136699938797392e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176700, "loss": 0.6262764539569616, "lr": 0.0003, "grad_norm": 0.13683514611344433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25860693896, "tokens_per_second": 148671.43800615135, "data_wait_s": 9.013001545099542e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 176800, "loss": 0.5874834768474102, "lr": 0.0003, "grad_norm": 0.132651790254468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25875330860, "tokens_per_second": 149868.69283462793, "data_wait_s": 1.1986998288193718e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 176900, "loss": 0.5908063501119614, "lr": 0.0003, "grad_norm": 0.1466829626245251, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25890070368, "tokens_per_second": 150322.62844061662, "data_wait_s": 1.2328000593697652e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4db3895d1d9e20ad9415a6dc95caae2aebe37cfcfc8417530c18e8f563b1c8b5
-size 465021
+oid sha256:d996c45a64d789ab97cae581fe47201f822d03eeca2aca39509a4d34028b15bc
+size 467139