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": 76526.58374068562,
-      "required_pages_per_second_upper_estimate": 0.005294286629318749
+      "required_token_positions_per_second": 77084.17372998579,
+      "required_pages_per_second_upper_estimate": 0.00533286199856556
     },
     {
       "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": 9565.822967585702,
-      "required_pages_per_second_upper_estimate": 0.004005010319429433
+      "required_token_positions_per_second": 9635.521716248224,
+      "required_pages_per_second_upper_estimate": 0.004034191730019052
     },
     {
       "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": 19131.645935171404,
-      "required_pages_per_second_upper_estimate": 0.003984617312864354
+      "required_token_positions_per_second": 19271.043432496448,
+      "required_pages_per_second_upper_estimate": 0.004013650135397946
     },
     {
       "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": 28697.468902757108,
-      "required_pages_per_second_upper_estimate": 0.0017203694422224475
+      "required_token_positions_per_second": 28906.56514874467,
+      "required_pages_per_second_upper_estimate": 0.0017329044428979206
     },
     {
       "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": 28697.468902757108,
-      "required_pages_per_second_upper_estimate": 0.002231101980363724
+      "required_token_positions_per_second": 28906.56514874467,
+      "required_pages_per_second_upper_estimate": 0.0022473582937720697
     },
     {
       "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": 28697.468902757108,
-      "required_pages_per_second_upper_estimate": 0.36810031814313704
+      "required_token_positions_per_second": 28906.56514874467,
+      "required_pages_per_second_upper_estimate": 0.3707823802766084
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12899730724237202,
-    "producer_p95_seconds": 0.2903622269986954,
-    "gpu_update_seconds": 0.9634299141045056,
-    "producer_batches_per_second": 7.752099802526172,
-    "gpu_batches_per_second": 1.0379582213092124,
-    "producer_headroom": 7.468604846877344,
+    "producer_batch_seconds": 0.12923500912495456,
+    "producer_p95_seconds": 0.2891170620023331,
+    "gpu_update_seconds": 0.9564609235905939,
+    "producer_batches_per_second": 7.737841369540366,
+    "gpu_batches_per_second": 1.0455210195581839,
+    "producer_headroom": 7.400942902908084,
     "sustainable": true,
-    "buffer_seconds": 60.69608458858385,
+    "buffer_seconds": 60.25703818620742,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339519.02450688067,
-    "upload_bytes_per_second": 5066127.074407039,
+    "generation_bytes_per_second": 327629.92646581755,
+    "upload_bytes_per_second": 4929080.673748696,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.85287906079115,
+    "estimated_full_shards_per_hour": 70.30175538521667,
     "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": 1.181815096999344,
-      "bytes": 13310041,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.608177831996727,
       "bytes": 5937731,
@@ -562,6 +556,12 @@
       "bytes": 12817447,
       "pages": 46,
       "shards": 4
+    },
+    {
+      "seconds": 1.0572433240013197,
+      "bytes": 864410,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 789800, "loss": 0.5162946581840515, "lr": 3e-05, "grad_norm": 0.25881686199596377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115604154353, "tokens_per_second": 152286.81515599234, "data_wait_s": 9.812996722757816e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 789900, "loss": 0.48603665828704834, "lr": 3e-05, "grad_norm": 0.24922216753557763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115618831669, "tokens_per_second": 152035.04025936045, "data_wait_s": 1.3707001926377416e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790000, "loss": 0.5306972861289978, "lr": 3e-05, "grad_norm": 0.2637444456124192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115633471654, "tokens_per_second": 152705.55062787078, "data_wait_s": 8.73099997988902e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 790100, "loss": 0.5542400479316711, "lr": 3e-05, "grad_norm": 0.260805359336468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115648197682, "tokens_per_second": 145252.18146208223, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d14a68185d17ecd922990237f078c3a0fff34a900e4547918dcfa9c3c0b73cf
-size 640263
+oid sha256:d15eb3fea9a2faf8e1fa306832696b47d6c836360cdf88c805906e3feae972ab
+size 642381