Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75104.60501539869,
-      "required_pages_per_second_upper_estimate": 0.006509876811853619
+      "required_token_positions_per_second": 75058.50693362516,
+      "required_pages_per_second_upper_estimate": 0.00650588114696001
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9388.075626924836,
-      "required_pages_per_second_upper_estimate": 0.0037450437318193855
+      "required_token_positions_per_second": 9382.313366703145,
+      "required_pages_per_second_upper_estimate": 0.0037427450800634854
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18776.15125384967,
-      "required_pages_per_second_upper_estimate": 0.004343532143602288
+      "required_token_positions_per_second": 18764.62673340629,
+      "required_pages_per_second_upper_estimate": 0.004340866148622344
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28164.226880774506,
-      "required_pages_per_second_upper_estimate": 0.0017150536507607038
+      "required_token_positions_per_second": 28146.940100109434,
+      "required_pages_per_second_upper_estimate": 0.0017140009765149295
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28164.226880774506,
-      "required_pages_per_second_upper_estimate": 0.002168160148851918
+      "required_token_positions_per_second": 28146.940100109434,
+      "required_pages_per_second_upper_estimate": 0.002166829364623449
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28164.226880774506,
-      "required_pages_per_second_upper_estimate": 0.349705438255392
+      "required_token_positions_per_second": 28146.940100109434,
+      "required_pages_per_second_upper_estimate": 0.349490794294665
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09525290772631934,
-    "producer_p95_seconds": 0.36458133300038753,
-    "gpu_update_seconds": 0.9816708307684138,
-    "producer_batches_per_second": 10.498367177128074,
-    "gpu_batches_per_second": 1.018671400490976,
-    "producer_headroom": 10.305940828483163,
+    "producer_batch_seconds": 0.13297151219506986,
+    "producer_p95_seconds": 0.3884939659983502,
+    "gpu_update_seconds": 0.9822737356766005,
+    "producer_batches_per_second": 7.520407818879243,
+    "gpu_batches_per_second": 1.0180461552412265,
+    "producer_headroom": 7.387099082062029,
     "sustainable": true,
-    "buffer_seconds": 60.86359150764165,
+    "buffer_seconds": 60.90097161194923,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339327.59582496714,
-    "upload_bytes_per_second": 5243837.951270891,
+    "generation_bytes_per_second": 337393.14615021425,
+    "upload_bytes_per_second": 5237578.227859753,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.81180292188417,
+    "estimated_full_shards_per_hour": 72.39671505336591,
     "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": 0.9746390660002362,
-      "bytes": 5704060,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 1.1505642880001687,
       "bytes": 6924827,
@@ -562,6 +556,12 @@
       "bytes": 7506339,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1381002670023008,
+      "bytes": 6063351,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 274400, "loss": 0.5805154629051685, "lr": 0.0003, "grad_norm": 0.14057968008857563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40163496188, "tokens_per_second": 148212.90861629412, "data_wait_s": 1.2228993000462651e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274500, "loss": 0.5555185135453939, "lr": 0.0003, "grad_norm": 0.14979814296883148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40178122704, "tokens_per_second": 148706.46688129418, "data_wait_s": 1.054699532687664e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274600, "loss": 0.551976615563035, "lr": 0.0003, "grad_norm": 0.15907131189507812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40192819820, "tokens_per_second": 150268.8566915282, "data_wait_s": 1.0700001439545304e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 274700, "loss": 0.5999009869992733, "lr": 0.0003, "grad_norm": 0.168274931024671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40207522222, "tokens_per_second": 150018.71773799387, "data_wait_s": 9.433999366592616e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af3e11fdae6dd5efaae9dc1922df70d108ade434427c4ab12d21a9db57531ec8
-size 799896
+oid sha256:8f3d36bea07e77adc8e7be1176731c0d93198c60ccd4ce526258aab30b597669
+size 802014