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": 77108.5678047361,
-      "required_pages_per_second_upper_estimate": 0.005334549637258837
+      "required_token_positions_per_second": 77040.85580878869,
+      "required_pages_per_second_upper_estimate": 0.005329865164265719
     },
     {
       "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": 9638.570975592012,
-      "required_pages_per_second_upper_estimate": 0.00403546839123052
+      "required_token_positions_per_second": 9630.106976098587,
+      "required_pages_per_second_upper_estimate": 0.0040319246913910305
     },
     {
       "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": 19277.141951184025,
-      "required_pages_per_second_upper_estimate": 0.004014920296011815
+      "required_token_positions_per_second": 19260.213952197173,
+      "required_pages_per_second_upper_estimate": 0.004011394640244208
     },
     {
       "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": 28915.712926776036,
-      "required_pages_per_second_upper_estimate": 0.0017334528382230527
+      "required_token_positions_per_second": 28890.320928295758,
+      "required_pages_per_second_upper_estimate": 0.0017319306266854967
     },
     {
       "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": 28915.712926776036,
-      "required_pages_per_second_upper_estimate": 0.0022480694932771853
+      "required_token_positions_per_second": 28890.320928295758,
+      "required_pages_per_second_upper_estimate": 0.0022460953770829413
     },
     {
       "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": 28915.712926776036,
-      "required_pages_per_second_upper_estimate": 0.370899718151076
+      "required_token_positions_per_second": 28890.320928295758,
+      "required_pages_per_second_upper_estimate": 0.3705740168583748
     }
   ],
   "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": 0.09170077097664375,
-    "producer_p95_seconds": 0.25694418599960045,
-    "gpu_update_seconds": 0.9561583375106021,
-    "producer_batches_per_second": 10.905033723813519,
-    "gpu_batches_per_second": 1.045851885372397,
-    "producer_headroom": 10.426938915858583,
+    "producer_batch_seconds": 0.12103978567969875,
+    "producer_p95_seconds": 0.2756980720005231,
+    "gpu_update_seconds": 0.956998714850585,
+    "producer_batches_per_second": 8.261746287673109,
+    "gpu_batches_per_second": 1.0449334826495862,
+    "producer_headroom": 7.906480579724757,
     "sustainable": true,
-    "buffer_seconds": 60.23797526316793,
+    "buffer_seconds": 60.29091903558685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306689.1680943889,
-    "upload_bytes_per_second": 4779566.796762844,
+    "generation_bytes_per_second": 305802.1646794354,
+    "upload_bytes_per_second": 4780459.348635522,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.80835611461401,
+    "estimated_full_shards_per_hour": 65.6180258301477,
     "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": 1.136130429997138,
-      "bytes": 965107,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.3451466709993838,
       "bytes": 7707115,
@@ -562,6 +556,12 @@
       "bytes": 12069614,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.154275818997121,
+      "bytes": 1133519,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"step": 912000, "loss": 0.554841160774231, "lr": 3e-05, "grad_norm": 0.32315140536455805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133490399358, "tokens_per_second": 153303.78350690423, "data_wait_s": 1.52729990077205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912100, "loss": 0.5340381860733032, "lr": 3e-05, "grad_norm": 0.33642669476817466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133505129415, "tokens_per_second": 153514.19093113282, "data_wait_s": 1.6370999219361693e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912200, "loss": 0.56758713722229, "lr": 3e-05, "grad_norm": 0.33254980483321406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133519863314, "tokens_per_second": 153573.38938487993, "data_wait_s": 1.5778998204041272e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 912300, "loss": 0.5520579218864441, "lr": 3e-05, "grad_norm": 0.32711840799363257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133534433724, "tokens_per_second": 150990.10566519367, "data_wait_s": 1.572600012877956e-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 sha256:440a9edb5ae87c7c7ed9c1989f6ab127f8c8bfdca99c196d0aca6a446d7c197b
-size 580176
+oid sha256:554a03aa523ef8c4071e1a38d57e27755d205fd1fd8e598dcb85bf7a8d6cd20c
+size 582294