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": 77008.47653304035,
-      "required_pages_per_second_upper_estimate": 0.005327625090839178
+      "required_token_positions_per_second": 76303.03742398215,
+      "required_pages_per_second_upper_estimate": 0.005278821176430286
     },
     {
       "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": 9626.059566630043,
-      "required_pages_per_second_upper_estimate": 0.004030230125566072
+      "required_token_positions_per_second": 9537.87967799777,
+      "required_pages_per_second_upper_estimate": 0.003993311047601207
     },
     {
       "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": 19252.119133260087,
-      "required_pages_per_second_upper_estimate": 0.004009708702934343
+      "required_token_positions_per_second": 19075.75935599554,
+      "required_pages_per_second_upper_estimate": 0.003972977612250225
     },
     {
       "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": 28878.17869989013,
-      "required_pages_per_second_upper_estimate": 0.0017312027186326886
+      "required_token_positions_per_second": 28613.639033993306,
+      "required_pages_per_second_upper_estimate": 0.0017153439695908556
     },
     {
       "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": 28878.17869989013,
-      "required_pages_per_second_upper_estimate": 0.0022451513722324223
+      "required_token_positions_per_second": 28613.639033993306,
+      "required_pages_per_second_upper_estimate": 0.002224584576795963
     },
     {
       "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": 28878.17869989013,
-      "required_pages_per_second_upper_estimate": 0.3704182693896965
+      "required_token_positions_per_second": 28613.639033993306,
+      "required_pages_per_second_upper_estimate": 0.36702503859613533
     }
   ],
   "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": 0.10904455229686505,
-    "producer_p95_seconds": 0.29280203099915525,
-    "gpu_update_seconds": 0.9574010981553068,
-    "producer_batches_per_second": 9.17056358099926,
-    "gpu_batches_per_second": 1.0444943106152391,
-    "producer_headroom": 8.779907643151756,
+    "producer_batch_seconds": 0.10518822914073667,
+    "producer_p95_seconds": 0.2929048750011134,
+    "gpu_update_seconds": 0.9662524912386671,
+    "producer_batches_per_second": 9.506767137053417,
+    "gpu_batches_per_second": 1.0349261803382996,
+    "producer_headroom": 9.185937429803756,
     "sustainable": true,
-    "buffer_seconds": 60.31626918378433,
+    "buffer_seconds": 60.87390694803602,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412777.7590896337,
-    "upload_bytes_per_second": 5416572.012920619,
+    "generation_bytes_per_second": 444322.73505109904,
+    "upload_bytes_per_second": 5318272.264420617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.57249812618979,
+    "estimated_full_shards_per_hour": 95.34131563806275,
     "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": 1.2517445409976062,
-      "bytes": 10940783,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2907915830001002,
       "bytes": 2619734,
@@ -562,6 +556,12 @@
       "bytes": 7565588,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9064518010018219,
+      "bytes": 1123131,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -284,3 +284,4 @@
 {"step": 673400, "loss": 0.5718695521354675, "lr": 0.0001, "grad_norm": 0.23013544544482514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98568308512, "tokens_per_second": 154093.95665601574, "data_wait_s": 1.2325999705353752e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673500, "loss": 0.6256731748580933, "lr": 0.0001, "grad_norm": 0.20255019155449147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98582989496, "tokens_per_second": 152949.60079562184, "data_wait_s": 1.3421999028651044e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673600, "loss": 0.5435763001441956, "lr": 0.0001, "grad_norm": 0.2397423740731034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98597483724, "tokens_per_second": 150260.20550584595, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 673700, "loss": 0.5673876404762268, "lr": 0.0001, "grad_norm": 0.2134560078421933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98611932000, "tokens_per_second": 150182.8946675715, "data_wait_s": 1.2245000107213855e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b02f1cd5ce5c3e8ea9e77822d50dc198beb70524ceb006bb79dbb6bc5e6f5790
-size 609828
+oid sha256:c5e75c0b50dc21207ac7bae5a8cae3c37bf63e8056593daa4f18f19ad6dcd982
+size 611946