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": 76501.08686152734,
-      "required_pages_per_second_upper_estimate": 0.005292522696057669
+      "required_token_positions_per_second": 76866.1296220272,
+      "required_pages_per_second_upper_estimate": 0.00531777720124495
     },
     {
       "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": 9562.635857690917,
-      "required_pages_per_second_upper_estimate": 0.0040036759430709095
+      "required_token_positions_per_second": 9608.2662027534,
+      "required_pages_per_second_upper_estimate": 0.004022780415678612
     },
     {
       "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": 19125.271715381834,
-      "required_pages_per_second_upper_estimate": 0.003983289730981668
+      "required_token_positions_per_second": 19216.5324055068,
+      "required_pages_per_second_upper_estimate": 0.004002296926028455
     },
     {
       "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": 28687.907573072753,
-      "required_pages_per_second_upper_estimate": 0.0017197962551071754
+      "required_token_positions_per_second": 28824.798608260204,
+      "required_pages_per_second_upper_estimate": 0.0017280026636461551
     },
     {
       "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": 28687.907573072753,
-      "required_pages_per_second_upper_estimate": 0.00223035862903661
+      "required_token_positions_per_second": 28824.798608260204,
+      "required_pages_per_second_upper_estimate": 0.002241001304902405
     },
     {
       "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": 28687.907573072753,
-      "required_pages_per_second_upper_estimate": 0.3679776756721021
+      "required_token_positions_per_second": 28824.798608260204,
+      "required_pages_per_second_upper_estimate": 0.3697335668893447
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3344143743830159,
-    "producer_p95_seconds": 0.5007852519993321,
-    "gpu_update_seconds": 0.9637510135438098,
-    "producer_batches_per_second": 2.9903020820949124,
-    "gpu_batches_per_second": 1.0376123977529208,
-    "producer_headroom": 2.8819066624211365,
+    "producer_batch_seconds": 0.09388111198444449,
+    "producer_p95_seconds": 0.32758199599629734,
+    "gpu_update_seconds": 0.9591740908842648,
+    "producer_batches_per_second": 10.651769870021285,
+    "gpu_batches_per_second": 1.0425636070695965,
+    "producer_headroom": 10.21690168138607,
     "sustainable": true,
-    "buffer_seconds": 60.71631385326002,
+    "buffer_seconds": 60.427967725708676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428180.8240366846,
-    "upload_bytes_per_second": 5823226.5728952065,
+    "generation_bytes_per_second": 427690.7499726939,
+    "upload_bytes_per_second": 5874521.577557032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.87763729882624,
+    "estimated_full_shards_per_hour": 91.77247881303418,
     "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": 1.0030813440025668,
-      "bytes": 5714103,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.237253362996853,
       "bytes": 8511115,
@@ -562,6 +556,12 @@
       "bytes": 2574706,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1363400170012028,
+      "bytes": 10297243,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 518200, "loss": 0.5854417085647583, "lr": 0.0003, "grad_norm": 0.18373835397396274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75846794304, "tokens_per_second": 152599.15833331883, "data_wait_s": 1.7915997887030244e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518300, "loss": 0.5540903210639954, "lr": 0.0003, "grad_norm": 0.16986225399070048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75861220870, "tokens_per_second": 150238.7015431016, "data_wait_s": 2.3802997020538896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518400, "loss": 0.5579666495323181, "lr": 0.0003, "grad_norm": 0.1459412654943725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75875780773, "tokens_per_second": 151963.10700100864, "data_wait_s": 1.7535996448714286e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 518500, "loss": 0.6236104965209961, "lr": 0.0003, "grad_norm": 0.16184169780102126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75890515189, "tokens_per_second": 154018.4040465675, "data_wait_s": 2.02369992621243e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76699e17a68652f5b317bead599cecadb7a2e4a0a982ba800b34135c415630ed
-size 712275
+oid sha256:42112f3b4664583f3641536d4bb3707c236801be93b87ea72e6980b730f5982b
+size 714393