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": 77453.817419227,
-      "required_pages_per_second_upper_estimate": 0.005358434806678792
+      "required_token_positions_per_second": 77632.26397443068,
+      "required_pages_per_second_upper_estimate": 0.005370780153420836
     },
     {
       "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": 9681.727177403374,
-      "required_pages_per_second_upper_estimate": 0.004053536991724964
+      "required_token_positions_per_second": 9704.032996803835,
+      "required_pages_per_second_upper_estimate": 0.00406287597250946
     },
     {
       "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": 19363.45435480675,
-      "required_pages_per_second_upper_estimate": 0.004032896893475277
+      "required_token_positions_per_second": 19408.06599360767,
+      "required_pages_per_second_upper_estimate": 0.004042188321349478
     },
     {
       "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": 29045.181532210125,
-      "required_pages_per_second_upper_estimate": 0.0017412142834317122
+      "required_token_positions_per_second": 29112.098990411505,
+      "required_pages_per_second_upper_estimate": 0.0017452258828738426
     },
     {
       "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": 29045.181532210125,
-      "required_pages_per_second_upper_estimate": 0.002258135107877474
+      "required_token_positions_per_second": 29112.098990411505,
+      "required_pages_per_second_upper_estimate": 0.002263337645913839
     },
     {
       "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": 29045.181532210125,
-      "required_pages_per_second_upper_estimate": 0.3725604024090266
+      "required_token_positions_per_second": 29112.098990411505,
+      "required_pages_per_second_upper_estimate": 0.37341874771246525
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2514586477966674,
-    "producer_p95_seconds": 0.21011276699937298,
-    "gpu_update_seconds": 0.9518962713088677,
-    "producer_batches_per_second": 3.9767970151840335,
-    "gpu_batches_per_second": 1.0505346329647758,
-    "producer_headroom": 3.785498250505916,
+    "producer_batch_seconds": 0.07668421131256764,
+    "producer_p95_seconds": 0.20827702399947157,
+    "gpu_update_seconds": 0.9497082298705521,
+    "producer_batches_per_second": 13.040494032389061,
+    "gpu_batches_per_second": 1.0529549692712494,
+    "producer_headroom": 12.384664504137714,
     "sustainable": true,
-    "buffer_seconds": 60.92136136376753,
+    "buffer_seconds": 60.78132671171534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460978.5619639881,
-    "upload_bytes_per_second": 5327136.264896274,
+    "generation_bytes_per_second": 460708.36208739865,
+    "upload_bytes_per_second": 5293284.760875263,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.91526836576207,
+    "estimated_full_shards_per_hour": 98.85728976217717,
     "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": 1.4537456819998624,
-      "bytes": 13399952,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.1661972729998524,
       "bytes": 5666265,
@@ -562,6 +556,12 @@
       "bytes": 927002,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.3852189129993349,
+      "bytes": 10288556,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -124,3 +124,4 @@
 {"step": 897400, "loss": 0.5102488994598389, "lr": 3e-05, "grad_norm": 0.3118720297717996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131354045000, "tokens_per_second": 152348.96066376314, "data_wait_s": 2.3840999347157776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897500, "loss": 0.5129820704460144, "lr": 3e-05, "grad_norm": 0.32767872062354386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131368784009, "tokens_per_second": 154032.81284392608, "data_wait_s": 1.7063999621313997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897600, "loss": 0.5346129536628723, "lr": 3e-05, "grad_norm": 0.316646003153802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131383490026, "tokens_per_second": 153774.59404020943, "data_wait_s": 1.1744999937945977e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 897700, "loss": 0.5432120561599731, "lr": 3e-05, "grad_norm": 0.32400115971118726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131398124685, "tokens_per_second": 152180.50494692393, "data_wait_s": 1.927499943121802e-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 sha256:77ed510cedfbe09a6363b94316c3a63a720c4ceef162c140aa397363d24e251b
-size 268599
+oid sha256:b3c78a5e0eb29ad8e36591e106a97d6bb905758927961d358c1c408c80bd7d89
+size 270717