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": 76840.8154058829,
-      "required_pages_per_second_upper_estimate": 0.0053160259050350125
+      "required_token_positions_per_second": 76405.58453369688,
+      "required_pages_per_second_upper_estimate": 0.005285915623422427
     },
     {
       "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": 9605.101925735362,
-      "required_pages_per_second_upper_estimate": 0.0040214555989687775
+      "required_token_positions_per_second": 9550.69806671211,
+      "required_pages_per_second_upper_estimate": 0.003998677839277506
     },
     {
       "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": 19210.203851470724,
-      "required_pages_per_second_upper_estimate": 0.004000978855117934
+      "required_token_positions_per_second": 19101.39613342422,
+      "required_pages_per_second_upper_estimate": 0.003978317076901334
     },
     {
       "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": 28815.305777206086,
-      "required_pages_per_second_upper_estimate": 0.0017274335829191745
+      "required_token_positions_per_second": 28652.094200136326,
+      "required_pages_per_second_upper_estimate": 0.0017176492980834907
     },
     {
       "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": 28815.305777206086,
-      "required_pages_per_second_upper_estimate": 0.002240263279042498
+      "required_token_positions_per_second": 28652.094200136326,
+      "required_pages_per_second_upper_estimate": 0.0022275742968171828
     },
     {
       "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": 28815.305777206086,
-      "required_pages_per_second_upper_estimate": 0.36961180304519037
+      "required_token_positions_per_second": 28652.094200136326,
+      "required_pages_per_second_upper_estimate": 0.36751830017747755
     }
   ],
   "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": 0.07516989693746723,
-    "producer_p95_seconds": 0.259656931997597,
-    "gpu_update_seconds": 0.9594900784245897,
-    "producer_batches_per_second": 13.303197699364757,
-    "gpu_batches_per_second": 1.0422202610389932,
-    "producer_headroom": 12.764286203861312,
+    "producer_batch_seconds": 0.23664724942952375,
+    "producer_p95_seconds": 0.26834070200129645,
+    "gpu_update_seconds": 0.9649556436216258,
+    "producer_batches_per_second": 4.225698808714916,
+    "gpu_batches_per_second": 1.0363170645303938,
+    "producer_headroom": 4.077611913714638,
     "sustainable": true,
-    "buffer_seconds": 60.44787494074915,
+    "buffer_seconds": 60.79220554816242,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380298.1331188356,
-    "upload_bytes_per_second": 4954915.103140554,
+    "generation_bytes_per_second": 369271.15677818283,
+    "upload_bytes_per_second": 4880378.2139911875,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.60312647985269,
+    "estimated_full_shards_per_hour": 79.23699405202021,
     "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": 1.1679330120005034,
-      "bytes": 10820666,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.4827527400011604,
       "bytes": 6647922,
@@ -562,6 +556,12 @@
       "bytes": 13248087,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.2414953469997272,
+      "bytes": 4717232,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 865500, "loss": 0.5028296709060669, "lr": 3e-05, "grad_norm": 0.31073990264405715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126685154588, "tokens_per_second": 152093.12917937845, "data_wait_s": 1.0292002116329968e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865600, "loss": 0.49452322721481323, "lr": 3e-05, "grad_norm": 0.3115898766344744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126699874619, "tokens_per_second": 152425.77580387867, "data_wait_s": 9.561001206748188e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 865700, "loss": 0.5129722952842712, "lr": 3e-05, "grad_norm": 0.30290459662874364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126714534807, "tokens_per_second": 151665.45865019006, "data_wait_s": 1.8237002223031595e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 865800, "loss": 0.5157815217971802, "lr": 3e-05, "grad_norm": 0.3087268007317333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126729082291, "tokens_per_second": 150663.84339931875, "data_wait_s": 1.9134000467602164e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de4876bc17be445be35ee7052d800c731e4d108f3ed2c4de6ccb1839d845f9e0
-size 495456
+oid sha256:2bcd7b7d497bc0c09c8f07e5b42b6ab04ef39d5595d0c73b3ac725bf76e3fd6d
+size 497574