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": 77102.12520365442,
-      "required_pages_per_second_upper_estimate": 0.005334103923167113
+      "required_token_positions_per_second": 77133.28800990662,
+      "required_pages_per_second_upper_estimate": 0.005336259838411313
     },
     {
       "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": 9637.765650456802,
-      "required_pages_per_second_upper_estimate": 0.004035131218413509
+      "required_token_positions_per_second": 9641.661001238328,
+      "required_pages_per_second_upper_estimate": 0.004036762120441559
     },
     {
       "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": 19275.531300913604,
-      "required_pages_per_second_upper_estimate": 0.004014584840036191
+      "required_token_positions_per_second": 19283.322002476656,
+      "required_pages_per_second_upper_estimate": 0.004016207437717158
     },
     {
       "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": 28913.29695137041,
-      "required_pages_per_second_upper_estimate": 0.001733308004186466
+      "required_token_positions_per_second": 28924.983003714984,
+      "required_pages_per_second_upper_estimate": 0.0017340085651809548
     },
     {
       "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": 28913.29695137041,
-      "required_pages_per_second_upper_estimate": 0.0022478816618161502
+      "required_token_positions_per_second": 28924.983003714984,
+      "required_pages_per_second_upper_estimate": 0.0022487902009844293
     },
     {
       "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": 28913.29695137041,
-      "required_pages_per_second_upper_estimate": 0.3708687286126449
+      "required_token_positions_per_second": 28924.983003714984,
+      "required_pages_per_second_upper_estimate": 0.37101862474461567
     }
   ],
   "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": 0.07139115582023692,
-    "producer_p95_seconds": 0.25618786699851626,
-    "gpu_update_seconds": 0.9562382334502175,
-    "producer_batches_per_second": 14.007337302648553,
-    "gpu_batches_per_second": 1.0457645020026913,
-    "producer_headroom": 13.394351477625989,
+    "producer_batch_seconds": 0.08758021772661095,
+    "producer_p95_seconds": 0.25984448099916335,
+    "gpu_update_seconds": 0.955851901328655,
+    "producer_batches_per_second": 11.418103607844234,
+    "gpu_batches_per_second": 1.0461871746135338,
+    "producer_headroom": 10.914016043125486,
     "sustainable": true,
-    "buffer_seconds": 60.24300870736371,
+    "buffer_seconds": 60.218669783705266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362425.56352324964,
-    "upload_bytes_per_second": 4694350.665801326,
+    "generation_bytes_per_second": 366453.77690604056,
+    "upload_bytes_per_second": 4699371.571888348,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.76808909676663,
+    "estimated_full_shards_per_hour": 78.63244991670526,
     "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": 1.3585308149995399,
-      "bytes": 7486305,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8041142309994029,
       "bytes": 1148050,
@@ -562,6 +556,12 @@
       "bytes": 7544891,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1021929470007308,
+      "bytes": 6736997,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -225,3 +225,4 @@
 {"step": 782500, "loss": 0.5395228266716003, "lr": 3e-05, "grad_norm": 0.27321719219650825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114535920061, "tokens_per_second": 152350.74587177663, "data_wait_s": 1.38870018417947e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782600, "loss": 0.5248761773109436, "lr": 3e-05, "grad_norm": 0.25657650341361304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114550402363, "tokens_per_second": 149781.29115750318, "data_wait_s": 1.0855997970793396e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782700, "loss": 0.49081888794898987, "lr": 3e-05, "grad_norm": 0.2465137265393661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114564978464, "tokens_per_second": 151430.1650988239, "data_wait_s": 1.4089000615058467e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 782800, "loss": 0.5230893492698669, "lr": 3e-05, "grad_norm": 0.2622827538015773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114579688632, "tokens_per_second": 152982.13293579072, "data_wait_s": 1.2935000995639712e-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:d51102048c8e82994c0709d23ae401dafa73ef2bcdb80dcc645a0d59554cc10e
-size 484083
+oid sha256:bdea14c0a2e67f35678d196946bc15dce1d3d359c886e695c8be8d1b4098a54b
+size 486201