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": 77529.00430932941,
-      "required_pages_per_second_upper_estimate": 0.005363636410193697
+      "required_token_positions_per_second": 77593.91179034056,
+      "required_pages_per_second_upper_estimate": 0.005368126860336154
     },
     {
       "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": 9691.125538666176,
-      "required_pages_per_second_upper_estimate": 0.004057471889325598
+      "required_token_positions_per_second": 9699.23897379257,
+      "required_pages_per_second_upper_estimate": 0.004060868815185228
     },
     {
       "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": 19382.251077332352,
-      "required_pages_per_second_upper_estimate": 0.004036811755074452
+      "required_token_positions_per_second": 19398.47794758514,
+      "required_pages_per_second_upper_estimate": 0.004040191384216762
     },
     {
       "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": 29073.376615998528,
-      "required_pages_per_second_upper_estimate": 0.0017429045356534268
+      "required_token_positions_per_second": 29097.716921377712,
+      "required_pages_per_second_upper_estimate": 0.0017443636997954146
     },
     {
       "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": 29073.376615998528,
-      "required_pages_per_second_upper_estimate": 0.002260327151624955
+      "required_token_positions_per_second": 29097.716921377712,
+      "required_pages_per_second_upper_estimate": 0.0022622195033064983
     },
     {
       "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": 29073.376615998528,
-      "required_pages_per_second_upper_estimate": 0.37292205867034195
+      "required_token_positions_per_second": 29097.716921377712,
+      "required_pages_per_second_upper_estimate": 0.3732342699731624
     }
   ],
   "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": 0.10459588350011018,
-    "producer_p95_seconds": 0.22604268900067837,
-    "gpu_update_seconds": 0.9509731313694685,
-    "producer_batches_per_second": 9.560605700117698,
-    "gpu_batches_per_second": 1.0515544204281875,
-    "producer_headroom": 9.091879140429716,
+    "producer_batch_seconds": 0.06070799000782756,
+    "producer_p95_seconds": 0.20285205400068662,
+    "gpu_update_seconds": 0.950177640215043,
+    "producer_batches_per_second": 16.472296313402275,
+    "gpu_batches_per_second": 1.0524347844827007,
+    "producer_headroom": 15.651607639991525,
     "sustainable": true,
-    "buffer_seconds": 60.86228040764598,
+    "buffer_seconds": 60.811368973762754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462132.8742986512,
-    "upload_bytes_per_second": 5378615.475166717,
+    "generation_bytes_per_second": 461543.8565637714,
+    "upload_bytes_per_second": 5325589.532773373,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.16295692176486,
+    "estimated_full_shards_per_hour": 99.03656742749077,
     "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": 1.56234888400013,
-      "bytes": 11134116,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.07002194100005,
       "bytes": 1145768,
@@ -562,6 +556,12 @@
       "bytes": 12688722,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.4358708510008,
+      "bytes": 6164418,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -120,3 +120,4 @@
 {"step": 897000, "loss": 0.5342614650726318, "lr": 3e-05, "grad_norm": 0.3248516682573976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131295500377, "tokens_per_second": 152090.90968864236, "data_wait_s": 1.5327999790315516e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897100, "loss": 0.4904569387435913, "lr": 3e-05, "grad_norm": 0.33091313297640074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131310127884, "tokens_per_second": 153058.75319743843, "data_wait_s": 1.304299985349644e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897200, "loss": 0.5151164531707764, "lr": 3e-05, "grad_norm": 0.31993874572475556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131324839262, "tokens_per_second": 153999.89372563612, "data_wait_s": 1.3525999747798778e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 897300, "loss": 0.4963752031326294, "lr": 3e-05, "grad_norm": 0.3112137685009082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131339475719, "tokens_per_second": 152511.61751028922, "data_wait_s": 1.7022000974975526e-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 sha256:7093c58727e630f8473419701f29b26a0bd39b998f7a75e2a7fb42229efefdec
-size 260127
+oid sha256:7c9dfe169c2bb6d63c9ce11f888be42632a856419a6c7948f936c1bcaf115c99
+size 262245