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": 77351.39362855525,
-      "required_pages_per_second_upper_estimate": 0.0053513488911841834
+      "required_token_positions_per_second": 77334.41418920546,
+      "required_pages_per_second_upper_estimate": 0.005350174214172238
     },
     {
       "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": 9668.924203569406,
-      "required_pages_per_second_upper_estimate": 0.0040481766539371775
+      "required_token_positions_per_second": 9666.801773650683,
+      "required_pages_per_second_upper_estimate": 0.004047288036851585
     },
     {
       "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": 19337.848407138812,
-      "required_pages_per_second_upper_estimate": 0.004027563849850296
+      "required_token_positions_per_second": 19333.603547301365,
+      "required_pages_per_second_upper_estimate": 0.004026679757490637
     },
     {
       "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": 29006.772610708216,
-      "required_pages_per_second_upper_estimate": 0.0017389117272347003
+      "required_token_positions_per_second": 29000.405320952046,
+      "required_pages_per_second_upper_estimate": 0.0017385300179361062
     },
     {
       "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": 29006.772610708216,
-      "required_pages_per_second_upper_estimate": 0.002255148983173692
+      "required_token_positions_per_second": 29000.405320952046,
+      "required_pages_per_second_upper_estimate": 0.0022546539544018975
     },
     {
       "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": 29006.772610708216,
-      "required_pages_per_second_upper_estimate": 0.37206773400428694
+      "required_token_positions_per_second": 29000.405320952046,
+      "required_pages_per_second_upper_estimate": 0.3719860612479579
     }
   ],
   "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": 0.0605552206250195,
-    "producer_p95_seconds": 0.20393209400026535,
-    "gpu_update_seconds": 0.9531567117464627,
-    "producer_batches_per_second": 16.513852805398443,
-    "gpu_batches_per_second": 1.0491454213942497,
-    "producer_headroom": 15.740289638258679,
+    "producer_batch_seconds": 0.07928650871090781,
+    "producer_p95_seconds": 0.20351010200101882,
+    "gpu_update_seconds": 0.9533659855445203,
+    "producer_batches_per_second": 12.61248623831037,
+    "gpu_batches_per_second": 1.0489151230089715,
+    "producer_headroom": 12.024315372753465,
     "sustainable": true,
-    "buffer_seconds": 60.048872840027144,
+    "buffer_seconds": 60.06205708930478,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351797.11384667555,
-    "upload_bytes_per_second": 3588932.441098353,
+    "generation_bytes_per_second": 366149.64533126407,
+    "upload_bytes_per_second": 3542193.4783308543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.4874712138195,
+    "estimated_full_shards_per_hour": 78.56719036057893,
     "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": 1.2744999360002112,
-      "bytes": 7432872,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.153623332999814,
       "bytes": 6057859,
@@ -562,6 +556,12 @@
       "bytes": 13319811,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1297152579991234,
+      "bytes": 1247239,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -155,3 +155,4 @@
 {"step": 940500, "loss": 0.5382846593856812, "lr": 3e-05, "grad_norm": 0.3356429533463776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137662492808, "tokens_per_second": 153278.8204821817, "data_wait_s": 1.438699837308377e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940600, "loss": 0.5275529623031616, "lr": 3e-05, "grad_norm": 0.3333261844303751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137677145445, "tokens_per_second": 152856.9717985757, "data_wait_s": 1.476199940952938e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940700, "loss": 0.5371867418289185, "lr": 3e-05, "grad_norm": 0.34323661647359505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137691947860, "tokens_per_second": 154600.14354784923, "data_wait_s": 1.632999919820577e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 940800, "loss": 0.5152949690818787, "lr": 3e-05, "grad_norm": 0.3460394793445464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137706454781, "tokens_per_second": 151085.31234444628, "data_wait_s": 1.1926000297535211e-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:c29e9e521fd8b1a4e8cd13e1cf5a98e8b79a2c04e38192bdead3981e1f483def
-size 335040
+oid sha256:7d195b4c00b821bbb17f6318b537ae6c337f02c81fa275f4116cf11a3ea111b4
+size 337158