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": 76608.29168155126,
-      "required_pages_per_second_upper_estimate": 0.005299939374256377
+      "required_token_positions_per_second": 77359.81711174686,
+      "required_pages_per_second_upper_estimate": 0.005351931647296553
     },
     {
       "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": 9576.036460193907,
-      "required_pages_per_second_upper_estimate": 0.0040092864955025105
+      "required_token_positions_per_second": 9669.977138968357,
+      "required_pages_per_second_upper_estimate": 0.004048617496000927
     },
     {
       "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": 19152.072920387815,
-      "required_pages_per_second_upper_estimate": 0.003988871715189107
+      "required_token_positions_per_second": 19339.954277936715,
+      "required_pages_per_second_upper_estimate": 0.004028002447201951
     },
     {
       "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": 28728.10938058172,
-      "required_pages_per_second_upper_estimate": 0.0017222062920827323
+      "required_token_positions_per_second": 29009.931416905074,
+      "required_pages_per_second_upper_estimate": 0.001739101092842984
     },
     {
       "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": 28728.10938058172,
-      "required_pages_per_second_upper_estimate": 0.002233484142741369
+      "required_token_positions_per_second": 29009.931416905074,
+      "required_pages_per_second_upper_estimate": 0.0022553945664613773
     },
     {
       "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": 28728.10938058172,
-      "required_pages_per_second_upper_estimate": 0.3684933412934893
+      "required_token_positions_per_second": 29009.931416905074,
+      "required_pages_per_second_upper_estimate": 0.3721082517785184
     }
   ],
   "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": 0.05133288711717654,
+    "producer_batch_seconds": 0.10705924866406491,
     "producer_p95_seconds": 0.18382348799968895,
-    "gpu_update_seconds": 0.9624023507334665,
-    "producer_batches_per_second": 19.480688816846015,
-    "gpu_batches_per_second": 1.0390664561842349,
-    "producer_headroom": 18.74826071123976,
+    "gpu_update_seconds": 0.9530529253126249,
+    "producer_batches_per_second": 9.340622248693737,
+    "gpu_batches_per_second": 1.0492596721970875,
+    "producer_headroom": 8.902107358357755,
     "sustainable": true,
-    "buffer_seconds": 60.631348096208384,
+    "buffer_seconds": 60.042334294695365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314815.50292168313,
-    "upload_bytes_per_second": 5167551.933893536,
+    "generation_bytes_per_second": 327854.99902186304,
+    "upload_bytes_per_second": 5161892.294772752,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.55207839715834,
+    "estimated_full_shards_per_hour": 70.35005071632307,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 5796244,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.0528361209999275,
+      "bytes": 5176693,
+      "pages": 44,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 890600, "loss": 0.5161193013191223, "lr": 3e-05, "grad_norm": 0.32479326522699076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130359127841, "tokens_per_second": 152914.48461358098, "data_wait_s": 1.3161999959265813e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890700, "loss": 0.5632334351539612, "lr": 3e-05, "grad_norm": 0.31817407662024405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130373755907, "tokens_per_second": 142281.99469000564, "data_wait_s": 9.66199968388537e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 890800, "loss": 0.5203015804290771, "lr": 3e-05, "grad_norm": 0.3147710947317427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130388333487, "tokens_per_second": 151918.67317398838, "data_wait_s": 1.0480999662831891e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 890900, "loss": 0.5388245582580566, "lr": 3e-05, "grad_norm": 0.3211956049093675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130402871994, "tokens_per_second": 151829.41128575586, "data_wait_s": 1.174099998024758e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13f6a19f56b36884be9259d5e6c788c6400be6784af9951675b153b91c725028
-size 123792
+oid sha256:5bc555bd0fc89f35f089e8ba48cf9bee4d106d5e02d996b47c8ea27b690e0c87
+size 125910