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": 76963.9324484391,
-      "required_pages_per_second_upper_estimate": 0.0053245434277099475
+      "required_token_positions_per_second": 76861.46050567979,
+      "required_pages_per_second_upper_estimate": 0.005317454180942194
     },
     {
       "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": 9620.491556054887,
-      "required_pages_per_second_upper_estimate": 0.004027898915811537
+      "required_token_positions_per_second": 9607.682563209974,
+      "required_pages_per_second_upper_estimate": 0.004022536057989559
     },
     {
       "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": 19240.983112109774,
-      "required_pages_per_second_upper_estimate": 0.00400738936340536
+      "required_token_positions_per_second": 19215.365126419947,
+      "required_pages_per_second_upper_estimate": 0.004002053812577884
     },
     {
       "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": 28861.474668164665,
-      "required_pages_per_second_upper_estimate": 0.0017302013374363267
+      "required_token_positions_per_second": 28823.047689629922,
+      "required_pages_per_second_upper_estimate": 0.0017278976987477677
     },
     {
       "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": 28861.474668164665,
-      "required_pages_per_second_upper_estimate": 0.002243852707239038
+      "required_token_positions_per_second": 28823.047689629922,
+      "required_pages_per_second_upper_estimate": 0.0022408651786803777
     },
     {
       "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": 28861.474668164665,
-      "required_pages_per_second_upper_estimate": 0.37020400800611414
+      "required_token_positions_per_second": 28823.047689629922,
+      "required_pages_per_second_upper_estimate": 0.36971110798514545
     }
   ],
   "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": 0.246251347984213,
-    "producer_p95_seconds": 0.28423090500291437,
-    "gpu_update_seconds": 0.9579552090765766,
-    "producer_batches_per_second": 4.060891476070658,
-    "gpu_batches_per_second": 1.0438901428010947,
-    "producer_headroom": 3.890152142996555,
+    "producer_batch_seconds": 0.09379945278917035,
+    "producer_p95_seconds": 0.26202814700081944,
+    "gpu_update_seconds": 0.9592323579975658,
+    "producer_batches_per_second": 10.661043004671509,
+    "gpu_batches_per_second": 1.0425002781260821,
+    "producer_headroom": 10.226417420084505,
     "sustainable": true,
-    "buffer_seconds": 60.35117817182433,
+    "buffer_seconds": 60.431638553846646,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399930.46979445056,
-    "upload_bytes_per_second": 4858789.560383992,
+    "generation_bytes_per_second": 401360.98505895026,
+    "upload_bytes_per_second": 4911805.032490403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.81576891303193,
+    "estimated_full_shards_per_hour": 86.12272418810254,
     "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": 1.0633560690002923,
-      "bytes": 6091672,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8510588639983325,
       "bytes": 1121057,
@@ -562,6 +556,12 @@
       "bytes": 7601822,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.273706002997642,
+      "bytes": 11750945,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -280,3 +280,4 @@
 {"step": 788000, "loss": 0.5319559574127197, "lr": 3e-05, "grad_norm": 0.2482103040428209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115340762931, "tokens_per_second": 151552.4247322025, "data_wait_s": 1.3317996490513906e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788100, "loss": 0.5543046593666077, "lr": 3e-05, "grad_norm": 0.2817639190031358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115355492805, "tokens_per_second": 153133.4143941105, "data_wait_s": 1.506800253991969e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788200, "loss": 0.5107440948486328, "lr": 3e-05, "grad_norm": 0.2539574485578654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115370260563, "tokens_per_second": 153330.8123883191, "data_wait_s": 2.0318002498243004e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 788300, "loss": 0.5174246430397034, "lr": 3e-05, "grad_norm": 0.2595425070103543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115384893956, "tokens_per_second": 152300.01003865947, "data_wait_s": 1.3340999430511147e-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 sha256:08ffa8a3380809968cb420238f0efaf13bd03482893e13dd9b0910ffa280ab1a
-size 601356
+oid sha256:03a176e72411f4fc3695bdbed4618696401c521038b95b74a5754438aa9ceef8
+size 603474