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": 76516.18979156492,
-      "required_pages_per_second_upper_estimate": 0.005293567551801295
+      "required_token_positions_per_second": 77142.93998967423,
+      "required_pages_per_second_upper_estimate": 0.0053369275847673136
     },
     {
       "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": 9564.523723945615,
-      "required_pages_per_second_upper_estimate": 0.004004466353248621
+      "required_token_positions_per_second": 9642.867498709278,
+      "required_pages_per_second_upper_estimate": 0.004037267255738114
     },
     {
       "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": 19129.04744789123,
-      "required_pages_per_second_upper_estimate": 0.003984076116490612
+      "required_token_positions_per_second": 19285.734997418556,
+      "required_pages_per_second_upper_estimate": 0.0040167100009286
     },
     {
       "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": 28693.571171836844,
-      "required_pages_per_second_upper_estimate": 0.001720135779205268
+      "required_token_positions_per_second": 28928.602496127834,
+      "required_pages_per_second_upper_estimate": 0.0017342255482244602
     },
     {
       "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": 28693.571171836844,
-      "required_pages_per_second_upper_estimate": 0.00223079894892898
+      "required_token_positions_per_second": 28928.602496127834,
+      "required_pages_per_second_upper_estimate": 0.0022490716005990636
     },
     {
       "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": 28693.571171836844,
-      "required_pages_per_second_upper_estimate": 0.3680503222359493
+      "required_token_positions_per_second": 28928.602496127834,
+      "required_pages_per_second_upper_estimate": 0.37106505170697957
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0949235066329095,
-    "producer_p95_seconds": 0.23455978900165064,
-    "gpu_update_seconds": 0.9635607862968592,
-    "producer_batches_per_second": 10.534798338911187,
-    "gpu_batches_per_second": 1.0378172443517377,
-    "producer_headroom": 10.15091857092011,
+    "producer_batch_seconds": 0.08498427919536766,
+    "producer_p95_seconds": 0.21850305199768627,
+    "gpu_update_seconds": 0.955732306933968,
+    "producer_batches_per_second": 11.766882174774134,
+    "gpu_batches_per_second": 1.0463180879675866,
+    "producer_headroom": 11.24598944631707,
     "sustainable": true,
-    "buffer_seconds": 60.704329536702126,
+    "buffer_seconds": 60.21113533683999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390809.800000064,
-    "upload_bytes_per_second": 5687264.127838366,
+    "generation_bytes_per_second": 390726.1986865885,
+    "upload_bytes_per_second": 5609262.739928213,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.85868549348298,
+    "estimated_full_shards_per_hour": 83.8407465977501,
     "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": 1.4440002779992938,
-      "bytes": 13477323,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 1.0930595740010176,
       "bytes": 6005049,
@@ -562,6 +556,12 @@
       "bytes": 2425262,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.180065475000447,
+      "bytes": 6094205,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -157,3 +157,4 @@
 {"step": 540700, "loss": 0.5719364881515503, "lr": 0.0003, "grad_norm": 0.15767005672336482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79140476289, "tokens_per_second": 154338.05389398625, "data_wait_s": 9.760999091668054e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 540800, "loss": 0.5983349680900574, "lr": 0.0003, "grad_norm": 0.15553669881718665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79155188870, "tokens_per_second": 153084.56724384284, "data_wait_s": 1.0845000360859558e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 540900, "loss": 0.6019031405448914, "lr": 0.0003, "grad_norm": 0.1659243901186051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79169914549, "tokens_per_second": 154091.6230715643, "data_wait_s": 1.4461998944170773e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 541000, "loss": 0.5442508459091187, "lr": 0.0003, "grad_norm": 0.15133859636974892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79184499378, "tokens_per_second": 152497.19862960867, "data_wait_s": 1.2751999747706577e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0bcf14cf1421db74230d63cb9a4de8596d4d025de71e0f98a4abcbb3d3d3893
-size 339276
+oid sha256:e85fc855d083de3b792649e722b848876dce2b39876ce8eaa194ab2112fc17f0
+size 341394