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": 77053.11526241472,
-      "required_pages_per_second_upper_estimate": 0.005330713301713419
+      "required_token_positions_per_second": 76781.99410840141,
+      "required_pages_per_second_upper_estimate": 0.00531195651119103
     },
     {
       "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": 9631.63940780184,
-      "required_pages_per_second_upper_estimate": 0.00403256628854437
+      "required_token_positions_per_second": 9597.749263550177,
+      "required_pages_per_second_upper_estimate": 0.0040183771928528865
     },
     {
       "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": 19263.27881560368,
-      "required_pages_per_second_upper_estimate": 0.0040120329704659
+      "required_token_positions_per_second": 19195.498527100353,
+      "required_pages_per_second_upper_estimate": 0.003997916123857068
     },
     {
       "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": 28894.91822340552,
-      "required_pages_per_second_upper_estimate": 0.0017322062275076638
+      "required_token_positions_per_second": 28793.247790650526,
+      "required_pages_per_second_upper_estimate": 0.001726111240305764
     },
     {
       "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": 28894.91822340552,
-      "required_pages_per_second_upper_estimate": 0.002246452796556361
+      "required_token_positions_per_second": 28793.247790650526,
+      "required_pages_per_second_upper_estimate": 0.0022385483676106324
     },
     {
       "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": 28894.91822340552,
-      "required_pages_per_second_upper_estimate": 0.3706329860238519
+      "required_token_positions_per_second": 28793.247790650526,
+      "required_pages_per_second_upper_estimate": 0.36932886687767635
     }
   ],
   "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": 0.3113639297343411,
-    "producer_p95_seconds": 0.2575442889974511,
-    "gpu_update_seconds": 0.9568464525919479,
-    "producer_batches_per_second": 3.211675806035754,
-    "gpu_batches_per_second": 1.045099762131276,
-    "producer_headroom": 3.0730806018806964,
+    "producer_batch_seconds": 0.10862986347649439,
+    "producer_p95_seconds": 0.26011653900059173,
+    "gpu_update_seconds": 0.9602251264262588,
+    "producer_batches_per_second": 9.205571727671208,
+    "gpu_batches_per_second": 1.0414224461317465,
+    "producer_headroom": 8.83942127602908,
     "sustainable": true,
-    "buffer_seconds": 60.28132651329272,
+    "buffer_seconds": 60.4941829648543,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291740.8901318381,
-    "upload_bytes_per_second": 4857058.4021261055,
+    "generation_bytes_per_second": 291872.34747734305,
+    "upload_bytes_per_second": 4930206.589925044,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.600803641952105,
+    "estimated_full_shards_per_hour": 62.629011328127085,
     "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": 1.149405411000771,
-      "bytes": 4840926,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.5049040789999708,
       "bytes": 13497203,
@@ -562,6 +556,12 @@
       "bytes": 900332,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.302292963002401,
+      "bytes": 11974442,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -252,3 +252,4 @@
 {"step": 785200, "loss": 0.4932396113872528, "lr": 3e-05, "grad_norm": 0.2637239050660287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114930938036, "tokens_per_second": 139444.15837518178, "data_wait_s": 9.243001841241494e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 785300, "loss": 0.5490309000015259, "lr": 3e-05, "grad_norm": 0.26077023025603313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114945582488, "tokens_per_second": 152146.7419252921, "data_wait_s": 1.3760000001639128e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 785400, "loss": 0.5478571653366089, "lr": 3e-05, "grad_norm": 0.2627609841850018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114960231236, "tokens_per_second": 151862.56571565307, "data_wait_s": 1.1579999409150332e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 785500, "loss": 0.5633331537246704, "lr": 3e-05, "grad_norm": 0.2577616388919846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114974904358, "tokens_per_second": 152468.21620589224, "data_wait_s": 1.5104997146409005e-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:e9e56bf836e6519d68d7090dc191877b5a8d270ddc6eb50329d7a098bcb2baed
-size 542052
+oid sha256:94491d1371d66b21f91f5cd611727e828fc9ee8e29f0a450d112ee4e94f56ccc
+size 544170