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": 75564.4390039296,
-      "required_pages_per_second_upper_estimate": 0.00654973405654089
+      "required_token_positions_per_second": 75579.05859252009,
+      "required_pages_per_second_upper_estimate": 0.006551001245426904
     },
     {
       "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": 9445.5548754912,
-      "required_pages_per_second_upper_estimate": 0.0037679730634638583
+      "required_token_positions_per_second": 9447.38232406501,
+      "required_pages_per_second_upper_estimate": 0.0037687020600227425
     },
     {
       "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": 18891.1097509824,
-      "required_pages_per_second_upper_estimate": 0.0043701257687134955
+      "required_token_positions_per_second": 18894.76464813002,
+      "required_pages_per_second_upper_estimate": 0.004370971264844605
     },
     {
       "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": 28336.6646264736,
-      "required_pages_per_second_upper_estimate": 0.0017255542047628467
+      "required_token_positions_per_second": 28342.14697219503,
+      "required_pages_per_second_upper_estimate": 0.0017258880508536367
     },
     {
       "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": 28336.6646264736,
-      "required_pages_per_second_upper_estimate": 0.0021814348838540653
+      "required_token_positions_per_second": 28342.14697219503,
+      "required_pages_per_second_upper_estimate": 0.002181856930003805
     },
     {
       "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": 28336.6646264736,
-      "required_pages_per_second_upper_estimate": 0.35184653794496445
+      "required_token_positions_per_second": 28342.14697219503,
+      "required_pages_per_second_upper_estimate": 0.3519146103305938
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1384991438908969,
-    "producer_p95_seconds": 0.3258870930003468,
-    "gpu_update_seconds": 0.9756970470748271,
-    "producer_batches_per_second": 7.220261237049615,
-    "gpu_batches_per_second": 1.0249082981218751,
-    "producer_headroom": 7.044787568098148,
+    "producer_batch_seconds": 0.37053423446104716,
+    "producer_p95_seconds": 0.3459888149991457,
+    "gpu_update_seconds": 0.9755083137182224,
+    "producer_batches_per_second": 2.698805959062134,
+    "gpu_batches_per_second": 1.0251065889827486,
+    "producer_headroom": 2.6327076501773923,
     "sustainable": true,
-    "buffer_seconds": 60.49321691863928,
+    "buffer_seconds": 60.48151545052979,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436731.69442555297,
-    "upload_bytes_per_second": 5216741.986453059,
+    "generation_bytes_per_second": 382187.6634426139,
+    "upload_bytes_per_second": 5322141.558102892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.7124550302023,
+    "estimated_full_shards_per_hour": 82.00857570132077,
     "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": 0.9203152180016332,
-      "bytes": 582089,
-      "pages": 18,
-      "shards": 1
-    },
     {
       "seconds": 1.208456311996997,
       "bytes": 13456782,
@@ -562,6 +556,12 @@
       "bytes": 8306730,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.168453648002469,
+      "bytes": 10300928,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 385800, "loss": 0.5800092741847038, "lr": 0.0003, "grad_norm": 0.14827453131526264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56469982706, "tokens_per_second": 148383.5879183736, "data_wait_s": 8.168997737811878e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 385900, "loss": 0.5632327254861593, "lr": 0.0003, "grad_norm": 0.1679229761758304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56484749011, "tokens_per_second": 150653.73762363667, "data_wait_s": 8.01799978944473e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386000, "loss": 0.5998760890215635, "lr": 0.0003, "grad_norm": 0.16240458667847157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56499428794, "tokens_per_second": 149967.0774129678, "data_wait_s": 1.3215001672506332e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 386100, "loss": 0.5972938779741526, "lr": 0.0003, "grad_norm": 0.16952418522300677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56514234255, "tokens_per_second": 151120.4774845238, "data_wait_s": 7.779999577905983e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:298518b7835ae699ea9c9f28f53efd52614adce1cd1aeed9cd8d27fe4faf2695
-size 661443
+oid sha256:cce64d39ff708d3058dbe3bf401a532d6bf0172b9b89fdcb1bb827dd698e9a32
+size 663561