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": 77116.54766106943,
-      "required_pages_per_second_upper_estimate": 0.005335101702235795
+      "required_token_positions_per_second": 76836.52373302281,
+      "required_pages_per_second_upper_estimate": 0.0053157289971485
     },
     {
       "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": 9639.568457633679,
-      "required_pages_per_second_upper_estimate": 0.004035886016131572
+      "required_token_positions_per_second": 9604.565466627851,
+      "required_pages_per_second_upper_estimate": 0.004021230994743003
     },
     {
       "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": 19279.136915267358,
-      "required_pages_per_second_upper_estimate": 0.004015335794419633
+      "required_token_positions_per_second": 19209.130933255703,
+      "required_pages_per_second_upper_estimate": 0.004000755394548501
     },
     {
       "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": 28918.705372901037,
-      "required_pages_per_second_upper_estimate": 0.0017336322307990458
+      "required_token_positions_per_second": 28813.696399883556,
+      "required_pages_per_second_upper_estimate": 0.0017273371032060657
     },
     {
       "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": 28918.705372901037,
-      "required_pages_per_second_upper_estimate": 0.0022483021428010247
+      "required_token_positions_per_second": 28813.696399883556,
+      "required_pages_per_second_upper_estimate": 0.002240138157034574
     },
     {
       "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": 28918.705372901037,
-      "required_pages_per_second_upper_estimate": 0.3709381020369292
+      "required_token_positions_per_second": 28813.696399883556,
+      "required_pages_per_second_upper_estimate": 0.36959115968091166
     }
   ],
   "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": 0.07450279828910311,
-    "producer_p95_seconds": 0.25050951700177393,
-    "gpu_update_seconds": 0.9560593962794828,
-    "producer_batches_per_second": 13.422314637358546,
-    "gpu_batches_per_second": 1.0459601191008767,
-    "producer_headroom": 12.832530028866277,
+    "producer_batch_seconds": 0.1453639095157655,
+    "producer_p95_seconds": 0.2483589479998045,
+    "gpu_update_seconds": 0.9595436703536494,
+    "producer_batches_per_second": 6.879286635391054,
+    "gpu_batches_per_second": 1.042162051500418,
+    "producer_headroom": 6.600975947537939,
     "sustainable": true,
-    "buffer_seconds": 60.23174196560741,
+    "buffer_seconds": 60.451251232279915,
     "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": 35252.52258949415,
-    "upload_bytes_per_second": 5138301.932374967,
+    "generation_bytes_per_second": 34720.99580760222,
+    "upload_bytes_per_second": 5190724.488005381,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 7.564370710979636,
+    "estimated_full_shards_per_hour": 7.450317436895848,
     "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": 1.1232291689993872,
-      "bytes": 903046,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.245589108999411,
       "bytes": 10344850,
@@ -562,6 +556,12 @@
       "bytes": 5409485,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2911368530003529,
+      "bytes": 5955841,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 710500, "loss": 0.5499357581138611, "lr": 8e-05, "grad_norm": 0.2216257295097479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103996979780, "tokens_per_second": 152426.13515339777, "data_wait_s": 2.030699761235155e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710600, "loss": 0.6709777116775513, "lr": 8e-05, "grad_norm": 0.33018890736848455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104011647925, "tokens_per_second": 72031.4540071703, "data_wait_s": 1.4995002857176587e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710700, "loss": 0.5578862428665161, "lr": 8e-05, "grad_norm": 0.20971052914242538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104026390979, "tokens_per_second": 153050.39375180213, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 710800, "loss": 0.5703843832015991, "lr": 8e-05, "grad_norm": 0.215808214121722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104041158263, "tokens_per_second": 154485.9335840067, "data_wait_s": 1.791400063666515e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af40614cd414289ab5a0efa637897dd68f0509f3040bae9853d4aa1c92478da6
-size 548406
+oid sha256:5fb8a730ab7f8e67e10d74015d3d1ef125439e03b26d6de22a25c1557d476291
+size 550524