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": 76405.69625805516,
-      "required_pages_per_second_upper_estimate": 0.00528592335277277
+      "required_token_positions_per_second": 76249.61529320957,
+      "required_pages_per_second_upper_estimate": 0.005275125309467017
     },
     {
       "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": 9550.712032256895,
-      "required_pages_per_second_upper_estimate": 0.003998683686359474
+      "required_token_positions_per_second": 9531.201911651196,
+      "required_pages_per_second_upper_estimate": 0.003990515206279515
     },
     {
       "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": 19101.42406451379,
-      "required_pages_per_second_upper_estimate": 0.0039783228942107
+      "required_token_positions_per_second": 19062.403823302393,
+      "required_pages_per_second_upper_estimate": 0.003970196006999325
     },
     {
       "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": 28652.136096770686,
-      "required_pages_per_second_upper_estimate": 0.0017176518097227482
+      "required_token_positions_per_second": 28593.605734953588,
+      "required_pages_per_second_upper_estimate": 0.001714143003902501
     },
     {
       "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": 28652.136096770686,
-      "required_pages_per_second_upper_estimate": 0.0022275775540962204
+      "required_token_positions_per_second": 28593.605734953588,
+      "required_pages_per_second_upper_estimate": 0.0022230270759127945
     },
     {
       "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": 28652.136096770686,
-      "required_pages_per_second_upper_estimate": 0.367518837582518
+      "required_token_positions_per_second": 28593.605734953588,
+      "required_pages_per_second_upper_estimate": 0.36676807294613445
     }
   ],
   "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": 0.14068972032816873,
-    "producer_p95_seconds": 0.29607343899988336,
-    "gpu_update_seconds": 0.9649542326136075,
-    "producer_batches_per_second": 7.107839845494249,
-    "gpu_batches_per_second": 1.036318579888986,
-    "producer_headroom": 6.858740143649325,
+    "producer_batch_seconds": 0.07762395034384895,
+    "producer_p95_seconds": 0.2729683910001768,
+    "gpu_update_seconds": 0.966929468646983,
+    "producer_batches_per_second": 12.882621865678365,
+    "gpu_batches_per_second": 1.034201596316319,
+    "producer_headroom": 12.456586715360386,
     "sustainable": true,
-    "buffer_seconds": 60.79211665465727,
+    "buffer_seconds": 60.91655652475993,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251108.9297798747,
-    "upload_bytes_per_second": 4683468.556343536,
+    "generation_bytes_per_second": 265700.6272989159,
+    "upload_bytes_per_second": 4679545.0310742315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.88213081404859,
+    "estimated_full_shards_per_hour": 57.01316942430122,
     "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": 1.087463227002445,
-      "bytes": 6477149,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7014945069968235,
       "bytes": 13448331,
@@ -562,6 +556,12 @@
       "bytes": 1064335,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.423822492997715,
+      "bytes": 7690011,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -294,3 +294,4 @@
 {"step": 914400, "loss": 0.5224106907844543, "lr": 3e-05, "grad_norm": 0.35033095834050637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133841463019, "tokens_per_second": 151270.9553593144, "data_wait_s": 1.5842000721022487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914500, "loss": 0.5214608907699585, "lr": 3e-05, "grad_norm": 0.3191385382969053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133856084120, "tokens_per_second": 151598.96451312004, "data_wait_s": 1.6672001947881654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914600, "loss": 0.4910697340965271, "lr": 3e-05, "grad_norm": 0.3203844987821864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133870868641, "tokens_per_second": 153707.61262290328, "data_wait_s": 9.931998647516593e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 914700, "loss": 0.6132705807685852, "lr": 3e-05, "grad_norm": 0.36856665212879103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133885605938, "tokens_per_second": 154593.4702946775, "data_wait_s": 1.4293000276666135e-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:b44825aba62d9b139efe6c3241a9dfc170a6cb1c7a3a747418d3fa636b6a3631
-size 631008
+oid sha256:7bd4890d43bce5a81f02622edf3c825d94ba0a4127d9cbac5446fc5e4a2c1e98
+size 633126