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": 77021.74248697858,
-      "required_pages_per_second_upper_estimate": 0.005328542860313879
+      "required_token_positions_per_second": 77100.96213323921,
+      "required_pages_per_second_upper_estimate": 0.0053340234592570935
     },
     {
       "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": 9627.717810872322,
-      "required_pages_per_second_upper_estimate": 0.004030924397802237
+      "required_token_positions_per_second": 9637.620266654902,
+      "required_pages_per_second_upper_estimate": 0.004035070349251612
     },
     {
       "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": 19255.435621744644,
-      "required_pages_per_second_upper_estimate": 0.004010399440023994
+      "required_token_positions_per_second": 19275.240533309803,
+      "required_pages_per_second_upper_estimate": 0.004014524280812376
     },
     {
       "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": 28883.153432616968,
-      "required_pages_per_second_upper_estimate": 0.0017315009462636854
+      "required_token_positions_per_second": 28912.860799964707,
+      "required_pages_per_second_upper_estimate": 0.001733281857575659
     },
     {
       "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": 28883.153432616968,
-      "required_pages_per_second_upper_estimate": 0.002245538135820397
+      "required_token_positions_per_second": 28912.860799964707,
+      "required_pages_per_second_upper_estimate": 0.0022478477529627844
     },
     {
       "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": 28883.153432616968,
-      "required_pages_per_second_upper_estimate": 0.37048207991966453
+      "required_token_positions_per_second": 28912.860799964707,
+      "required_pages_per_second_upper_estimate": 0.3708631341307154
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05520107451557976,
-    "producer_p95_seconds": 0.17682927599980758,
-    "gpu_update_seconds": 0.9572361987586112,
-    "producer_batches_per_second": 18.115589393423193,
-    "gpu_batches_per_second": 1.0446742416311114,
-    "producer_headroom": 17.34089792923223,
+    "producer_batch_seconds": 0.07010453023434593,
+    "producer_p95_seconds": 0.17464263700003357,
+    "gpu_update_seconds": 0.9562526583337526,
+    "producer_batches_per_second": 14.26441339321714,
+    "gpu_batches_per_second": 1.0457487268505754,
+    "producer_headroom": 13.640383226835475,
     "sustainable": true,
-    "buffer_seconds": 60.30588052179251,
+    "buffer_seconds": 60.243917475026414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453703.4617403417,
-    "upload_bytes_per_second": 4660002.022318828,
+    "generation_bytes_per_second": 456176.54320880823,
+    "upload_bytes_per_second": 4718197.056527261,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.35420121343314,
+    "estimated_full_shards_per_hour": 97.88486692617593,
     "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": [
@@ -274,6 +274,12 @@
       "bytes": 921426,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.0380561770007262,
+      "bytes": 6122044,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -53,3 +53,4 @@
 {"step": 765300, "loss": 0.532582700252533, "lr": 3e-05, "grad_norm": 0.23488419984191805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112017439376, "tokens_per_second": 150846.18801539313, "data_wait_s": 1.6494999727001414e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765400, "loss": 0.4953550100326538, "lr": 3e-05, "grad_norm": 0.2328708429836473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112031963356, "tokens_per_second": 150755.27137500362, "data_wait_s": 1.7504999959783163e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765500, "loss": 0.5054397583007812, "lr": 3e-05, "grad_norm": 0.25784802535966006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112046612366, "tokens_per_second": 152478.54749228052, "data_wait_s": 1.0666999514796771e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 765600, "loss": 0.5014462471008301, "lr": 3e-05, "grad_norm": 0.23670771429161894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112061341022, "tokens_per_second": 153356.97703893794, "data_wait_s": 1.1054000424337573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4bd0ef3aeb1b138b80c2911b6ebc1d4fbf4c3cb256f1daf8cb510ab77f7ba4a
-size 117438
+oid sha256:38401487cf105e9cc91d2bffb7f137851794d602534ed1fa421dd9a0118e4536
+size 119556