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": 76433.77138520507,
-      "required_pages_per_second_upper_estimate": 0.00528786565521227
+      "required_token_positions_per_second": 76485.27707378077,
+      "required_pages_per_second_upper_estimate": 0.0052914289382576
     },
     {
       "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": 9554.221423150633,
-      "required_pages_per_second_upper_estimate": 0.004000152995042267
+      "required_token_positions_per_second": 9560.659634222597,
+      "required_pages_per_second_upper_estimate": 0.0040028485395729625
     },
     {
       "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": 19108.442846301266,
-      "required_pages_per_second_upper_estimate": 0.0039797847213593075
+      "required_token_positions_per_second": 19121.319268445193,
+      "required_pages_per_second_upper_estimate": 0.0039824665405178
     },
     {
       "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": 28662.6642694519,
-      "required_pages_per_second_upper_estimate": 0.0017182829575993982
+      "required_token_positions_per_second": 28681.978902667794,
+      "required_pages_per_second_upper_estimate": 0.0017194408403689537
     },
     {
       "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": 28662.6642694519,
-      "required_pages_per_second_upper_estimate": 0.002228396073213647
+      "required_token_positions_per_second": 28681.978902667794,
+      "required_pages_per_second_upper_estimate": 0.0022298977009900895
     },
     {
       "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": 28662.6642694519,
-      "required_pages_per_second_upper_estimate": 0.36765388167740154
+      "required_token_positions_per_second": 28681.978902667794,
+      "required_pages_per_second_upper_estimate": 0.3679016290538576
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0565937064763915,
-    "producer_p95_seconds": 0.19485348499983957,
-    "gpu_update_seconds": 0.9645997922623925,
-    "producer_batches_per_second": 17.669809282011908,
-    "gpu_batches_per_second": 1.0366993731717267,
-    "producer_headroom": 17.04429436274478,
+    "producer_batch_seconds": 0.07909351641403362,
+    "producer_p95_seconds": 0.19428231900019455,
+    "gpu_update_seconds": 0.9639502244187336,
+    "producer_batches_per_second": 12.64326136121278,
+    "gpu_batches_per_second": 1.0373979637828339,
+    "producer_headroom": 12.187474626525763,
     "sustainable": true,
-    "buffer_seconds": 60.769786912530726,
+    "buffer_seconds": 60.72886413838022,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365870.402532381,
-    "upload_bytes_per_second": 4120926.539327784,
+    "generation_bytes_per_second": 365263.85848929564,
+    "upload_bytes_per_second": 4050430.6742543913,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.5072713563783,
+    "estimated_full_shards_per_hour": 78.37712112435486,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 7430334,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.6944532489997073,
+      "bytes": 1160058,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 810100, "loss": 0.5151100754737854, "lr": 3e-05, "grad_norm": 0.27117276915334454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118575583382, "tokens_per_second": 146496.49573108696, "data_wait_s": 2.8355998438200913e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 810200, "loss": 0.4969123899936676, "lr": 3e-05, "grad_norm": 0.2838330049893852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118590098154, "tokens_per_second": 150436.83683852424, "data_wait_s": 1.5110999811440706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 810300, "loss": 0.5421403050422668, "lr": 3e-05, "grad_norm": 0.28415872772744255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118604795733, "tokens_per_second": 152200.08156263598, "data_wait_s": 1.5181998605839908e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 810400, "loss": 0.5186997056007385, "lr": 3e-05, "grad_norm": 0.26816050786455403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118619442095, "tokens_per_second": 151912.84056032586, "data_wait_s": 1.0404000931885093e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6f9881b9fe411ec1e26d5c1aa974e4c247801d34e6e7804bb5ad00e5ee409f8
-size 219885
+oid sha256:5ec63803ee9a5e22e697697b02d7018d7563de318288095d02ba90616722bfa7
+size 222003