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": 76915.34853099794,
-      "required_pages_per_second_upper_estimate": 0.005321182279571149
+      "required_token_positions_per_second": 76998.94617318414,
+      "required_pages_per_second_upper_estimate": 0.0053269657584308445
     },
     {
       "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": 9614.418566374743,
-      "required_pages_per_second_upper_estimate": 0.004025356281850906
+      "required_token_positions_per_second": 9624.868271648018,
+      "required_pages_per_second_upper_estimate": 0.004029731355234167
     },
     {
       "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": 19228.837132749486,
-      "required_pages_per_second_upper_estimate": 0.004004859676215628
+      "required_token_positions_per_second": 19249.736543296036,
+      "required_pages_per_second_upper_estimate": 0.0040092124722779544
     },
     {
       "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": 28843.255699124227,
-      "required_pages_per_second_upper_estimate": 0.0017291091380611074
+      "required_token_positions_per_second": 28874.604814944054,
+      "required_pages_per_second_upper_estimate": 0.0017309884696871778
     },
     {
       "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": 28843.255699124227,
-      "required_pages_per_second_upper_estimate": 0.0022424362625329194
+      "required_token_positions_per_second": 28874.604814944054,
+      "required_pages_per_second_upper_estimate": 0.0022448735183977233
     },
     {
       "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": 28843.255699124227,
-      "required_pages_per_second_upper_estimate": 0.36997031463326824
+      "required_token_positions_per_second": 28874.604814944054,
+      "required_pages_per_second_upper_estimate": 0.37037242743094695
     }
   ],
   "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": 0.3773386661327862,
-    "producer_p95_seconds": 0.2602806180002517,
-    "gpu_update_seconds": 0.9585603056883583,
-    "producer_batches_per_second": 2.6501392243966273,
-    "gpu_batches_per_second": 1.0432311812472594,
-    "producer_headroom": 2.54031826505434,
+    "producer_batch_seconds": 0.10898986739069016,
+    "producer_p95_seconds": 0.2410195440006646,
+    "gpu_update_seconds": 0.9575195981796009,
+    "producer_batches_per_second": 9.175164847346345,
+    "gpu_batches_per_second": 1.0443650468368075,
+    "producer_headroom": 8.78540015786267,
     "sustainable": true,
-    "buffer_seconds": 60.38929925836657,
+    "buffer_seconds": 60.323734685314854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361299.4037673919,
-    "upload_bytes_per_second": 5603673.832901362,
+    "generation_bytes_per_second": 430038.694905731,
+    "upload_bytes_per_second": 5504252.0436081635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.52644142881697,
+    "estimated_full_shards_per_hour": 92.27629313830326,
     "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": 1.3156409709999934,
-      "bytes": 11593993,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 2.0578567419997853,
       "bytes": 11618530,
@@ -562,6 +556,12 @@
       "bytes": 12112969,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0779387669990683,
+      "bytes": 2658914,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 411300, "loss": 0.5872836112976074, "lr": 0.0003, "grad_norm": 0.16727288171828153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60203851133, "tokens_per_second": 153061.57260279963, "data_wait_s": 1.3950999345979653e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411400, "loss": 0.5916213393211365, "lr": 0.0003, "grad_norm": 0.15783162479039536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60218561936, "tokens_per_second": 152024.59177569614, "data_wait_s": 1.2940998203703202e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411500, "loss": 0.6054062843322754, "lr": 0.0003, "grad_norm": 0.17300995907106745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60233332164, "tokens_per_second": 153308.91245792247, "data_wait_s": 1.547600004414562e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 411600, "loss": 0.6161276698112488, "lr": 0.0003, "grad_norm": 0.16131318474121076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60248041387, "tokens_per_second": 152322.1890086199, "data_wait_s": 1.2078000509063713e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:145d7a03f21a76ba02a227e41b4b36c78c0a7cf041351d58bfee557a5456ebc1
-size 307506
+oid sha256:8742f88bbe156c2071508509b2db1f78fd9ceba466cd74b2aa17e6f584982eec
+size 309624