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": 76259.52722685967,
-      "required_pages_per_second_upper_estimate": 0.005275811040035786
+      "required_token_positions_per_second": 77081.48527353721,
+      "required_pages_per_second_upper_estimate": 0.005332676004910368
     },
     {
       "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": 9532.44090335746,
-      "required_pages_per_second_upper_estimate": 0.003991033946233838
+      "required_token_positions_per_second": 9635.185659192151,
+      "required_pages_per_second_upper_estimate": 0.0040340510299473435
     },
     {
       "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": 19064.88180671492,
-      "required_pages_per_second_upper_estimate": 0.0039707121055953375
+      "required_token_positions_per_second": 19270.371318384303,
+      "required_pages_per_second_upper_estimate": 0.004013510151753227
     },
     {
       "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": 28597.32271007238,
-      "required_pages_per_second_upper_estimate": 0.0017143658308853803
+      "required_token_positions_per_second": 28905.556977576452,
+      "required_pages_per_second_upper_estimate": 0.001732844004575774
     },
     {
       "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": 28597.32271007238,
-      "required_pages_per_second_upper_estimate": 0.002223316054379035
+      "required_token_positions_per_second": 28905.556977576452,
+      "required_pages_per_second_upper_estimate": 0.002247279912898217
     },
     {
       "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": 28597.32271007238,
-      "required_pages_per_second_upper_estimate": 0.3668157503119814
+      "required_token_positions_per_second": 28905.556977576452,
+      "required_pages_per_second_upper_estimate": 0.3707694485393524
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06081880669535167,
-    "producer_p95_seconds": 0.20806680699752178,
-    "gpu_update_seconds": 0.9668037907010779,
-    "producer_batches_per_second": 16.442282483592844,
-    "gpu_batches_per_second": 1.0343360355205578,
-    "producer_headroom": 15.896461032915497,
+    "producer_batch_seconds": 0.3464049459297769,
+    "producer_p95_seconds": 0.23339492499871994,
+    "gpu_update_seconds": 0.9564942831389824,
+    "producer_batches_per_second": 2.8867948098025127,
+    "gpu_batches_per_second": 1.0454845550338705,
+    "producer_headroom": 2.7612027321713897,
     "sustainable": true,
-    "buffer_seconds": 60.90863881416791,
+    "buffer_seconds": 60.259139837755896,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416734.53621383774,
-    "upload_bytes_per_second": 5233706.579858885,
+    "generation_bytes_per_second": 383824.5324447488,
+    "upload_bytes_per_second": 5156384.89242927,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.42153038798665,
+    "estimated_full_shards_per_hour": 82.35980968481873,
     "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": 1.0591198959991743,
-      "bytes": 7189815,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.272207456000615,
       "bytes": 11979936,
@@ -562,6 +556,12 @@
       "bytes": 7877906,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.147284323000349,
+      "bytes": 1234396,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 623100, "loss": 0.557693362236023, "lr": 0.0001, "grad_norm": 0.15968168327072585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91205190000, "tokens_per_second": 153207.82871393726, "data_wait_s": 8.50900323712267e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 623200, "loss": 0.5366798043251038, "lr": 0.0001, "grad_norm": 0.1496097085517918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91219841701, "tokens_per_second": 152678.1553690973, "data_wait_s": 9.409999620402232e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 623300, "loss": 0.5710268616676331, "lr": 0.0001, "grad_norm": 0.1610284059102318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91234490323, "tokens_per_second": 152160.3451231219, "data_wait_s": 1.5248999261530116e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 623400, "loss": 0.5163668990135193, "lr": 0.0001, "grad_norm": 0.15465748857237932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91248986456, "tokens_per_second": 150213.52096789624, "data_wait_s": 1.3701002899324521e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07133542a4480c9d7fb912eaca41ec48d36ab224b66b6c750891e41bb31e9d6f
-size 403599
+oid sha256:6df498e8f7baa3d07a0648143f39b7dfee045162febfc8deedee1e9596aad843
+size 405717