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": 75111.75740261882,
-      "required_pages_per_second_upper_estimate": 0.006510496762650301
+      "required_token_positions_per_second": 75210.2159865413,
+      "required_pages_per_second_upper_estimate": 0.006519030903163701
     },
     {
       "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": 9388.969675327353,
-      "required_pages_per_second_upper_estimate": 0.0037454003810943647
+      "required_token_positions_per_second": 9401.276998317662,
+      "required_pages_per_second_upper_estimate": 0.0037503099562460757
     },
     {
       "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": 18777.939350654706,
-      "required_pages_per_second_upper_estimate": 0.004343945788328704
+      "required_token_positions_per_second": 18802.553996635324,
+      "required_pages_per_second_upper_estimate": 0.004349639953473346
     },
     {
       "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": 28166.90902598206,
-      "required_pages_per_second_upper_estimate": 0.0017152169793317154
+      "required_token_positions_per_second": 28203.830994952983,
+      "required_pages_per_second_upper_estimate": 0.0017174653335274444
     },
     {
       "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": 28166.90902598206,
-      "required_pages_per_second_upper_estimate": 0.0021683666278146485
+      "required_token_positions_per_second": 28203.830994952983,
+      "required_pages_per_second_upper_estimate": 0.002171208983192582
     },
     {
       "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": 28166.90902598206,
-      "required_pages_per_second_upper_estimate": 0.3497387415223073
+      "required_token_positions_per_second": 28203.830994952983,
+      "required_pages_per_second_upper_estimate": 0.350197188807045
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09879725060938682,
-    "producer_p95_seconds": 0.19439383700046164,
-    "gpu_update_seconds": 0.9815773528609972,
-    "producer_batches_per_second": 10.121739156018467,
-    "gpu_batches_per_second": 1.0187684109513186,
-    "producer_headroom": 9.935269927114112,
+    "producer_batch_seconds": 0.08898600646095645,
+    "producer_p95_seconds": 0.20540237800014438,
+    "gpu_update_seconds": 0.9802923583305952,
+    "producer_batches_per_second": 11.237721971923312,
+    "gpu_batches_per_second": 1.0201038409632879,
+    "producer_headroom": 11.01625297412025,
     "sustainable": true,
-    "buffer_seconds": 60.857795877381825,
+    "buffer_seconds": 60.77812621649691,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448849.6455962798,
-    "upload_bytes_per_second": 5435838.021156823,
+    "generation_bytes_per_second": 554814.518019056,
+    "upload_bytes_per_second": 5417461.619547132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.31268525997444,
+    "estimated_full_shards_per_hour": 119.0502801459194,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 15806866,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.6009747379994224,
+      "bytes": 7396060,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 322100, "loss": 0.5703791789710522, "lr": 0.0003, "grad_norm": 0.1460847661649317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47146211947, "tokens_per_second": 148092.0903678938, "data_wait_s": 1.1665999409160577e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322200, "loss": 0.6169965155422688, "lr": 0.0003, "grad_norm": 0.15325261118740036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47160713306, "tokens_per_second": 147730.26352134856, "data_wait_s": 1.538400010758778e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322300, "loss": 0.5765353012830019, "lr": 0.0003, "grad_norm": 0.157324963402977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47175254349, "tokens_per_second": 147989.39596911386, "data_wait_s": 1.6527000298083294e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 322400, "loss": 0.5849626809358597, "lr": 0.0003, "grad_norm": 0.1463768283313687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47189920207, "tokens_per_second": 149146.89040614478, "data_wait_s": 1.9225999494665302e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8d8a97c8000f315b0815d6a8f877041e60a256a31f261a1e0a4e923a6fcce35
-size 155562
+oid sha256:59352f0df98085624335a23e58d5b2816d441fb4bbaae5be695636d0b5762b78
+size 157680