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": 76970.14467561392,
-      "required_pages_per_second_upper_estimate": 0.005324973204000254
+      "required_token_positions_per_second": 76977.06568538626,
+      "required_pages_per_second_upper_estimate": 0.005325452015515261
     },
     {
       "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": 9621.26808445174,
-      "required_pages_per_second_upper_estimate": 0.004028224032035542
+      "required_token_positions_per_second": 9622.133210673283,
+      "required_pages_per_second_upper_estimate": 0.004028586242318613
     },
     {
       "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": 19242.53616890348,
-      "required_pages_per_second_upper_estimate": 0.004007712824178627
+      "required_token_positions_per_second": 19244.266421346565,
+      "required_pages_per_second_upper_estimate": 0.004008073190132696
     },
     {
       "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": 28863.80425335522,
-      "required_pages_per_second_upper_estimate": 0.00173034099251143
+      "required_token_positions_per_second": 28866.399632019846,
+      "required_pages_per_second_upper_estimate": 0.0017304965814994603
     },
     {
       "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": 28863.80425335522,
-      "required_pages_per_second_upper_estimate": 0.0022440338222408413
+      "required_token_positions_per_second": 28866.399632019846,
+      "required_pages_per_second_upper_estimate": 0.002244235601516152
     },
     {
       "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": 28863.80425335522,
-      "required_pages_per_second_upper_estimate": 0.3702338894236249
+      "required_token_positions_per_second": 28866.399632019846,
+      "required_pages_per_second_upper_estimate": 0.3702671801544342
     }
   ],
   "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": 0.05244641853124676,
-    "producer_p95_seconds": 0.17969072899995808,
-    "gpu_update_seconds": 0.9578778929248769,
-    "producer_batches_per_second": 19.067078896993042,
-    "gpu_batches_per_second": 1.0439744015247114,
-    "producer_headroom": 18.26393335808408,
+    "producer_batch_seconds": 0.077113785171818,
+    "producer_p95_seconds": 0.1734719550004229,
+    "gpu_update_seconds": 0.9577917701011682,
+    "producer_batches_per_second": 12.967850012444467,
+    "gpu_batches_per_second": 1.0440682737275697,
+    "producer_headroom": 12.420500017825644,
     "sustainable": true,
-    "buffer_seconds": 60.346307254267245,
+    "buffer_seconds": 60.3408815163736,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337655.43159849517,
-    "upload_bytes_per_second": 4070459.5056393086,
+    "generation_bytes_per_second": 337359.62289306254,
+    "upload_bytes_per_second": 4030107.7064312724,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.45299540487424,
+    "estimated_full_shards_per_hour": 72.38952174276264,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 17435942,
       "pages": 43,
       "shards": 5
+    },
+    {
+      "seconds": 2.0315385989997594,
+      "bytes": 5979371,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 933000, "loss": 0.5141292810440063, "lr": 3e-05, "grad_norm": 0.346287447739729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136564981385, "tokens_per_second": 152895.9097757887, "data_wait_s": 1.5439999515365344e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 933100, "loss": 0.5349289178848267, "lr": 3e-05, "grad_norm": 0.33806156412260024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136579669358, "tokens_per_second": 153346.9033480814, "data_wait_s": 7.098000423866324e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 933200, "loss": 0.5128114223480225, "lr": 3e-05, "grad_norm": 0.34213086251517383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136594359750, "tokens_per_second": 153478.80708891235, "data_wait_s": 1.0372999895480461e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 933300, "loss": 0.47830837965011597, "lr": 3e-05, "grad_norm": 0.33232681675727915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136608948932, "tokens_per_second": 152243.88730087108, "data_wait_s": 1.3970000509289093e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c30afd4c4ac1153dcc96ac13863c3b0846086095b27afd684aa2f8901e29da35
-size 174624
+oid sha256:e6eb567d0135ba38314596f6a9aaa549bb2f83f0c369085777c97685f576c829
+size 176742