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": 77351.82585064945,
-      "required_pages_per_second_upper_estimate": 0.005351378793311553
+      "required_token_positions_per_second": 77333.89235156549,
+      "required_pages_per_second_upper_estimate": 0.005350138112233474
     },
     {
       "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": 9668.97823133118,
-      "required_pages_per_second_upper_estimate": 0.004048199274232805
+      "required_token_positions_per_second": 9666.736543945686,
+      "required_pages_per_second_upper_estimate": 0.0040472607265362535
     },
     {
       "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": 19337.95646266236,
-      "required_pages_per_second_upper_estimate": 0.004027586354966235
+      "required_token_positions_per_second": 19333.473087891372,
+      "required_pages_per_second_upper_estimate": 0.004026652586235982
     },
     {
       "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": 29006.934693993542,
-      "required_pages_per_second_upper_estimate": 0.0017389214438801686
+      "required_token_positions_per_second": 29000.20963183706,
+      "required_pages_per_second_upper_estimate": 0.001738518286672463
     },
     {
       "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": 29006.934693993542,
-      "required_pages_per_second_upper_estimate": 0.002255161584436197
+      "required_token_positions_per_second": 29000.20963183706,
+      "required_pages_per_second_upper_estimate": 0.0022546387404339526
     },
     {
       "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": 29006.934693993542,
-      "required_pages_per_second_upper_estimate": 0.37206981303463965
+      "required_token_positions_per_second": 29000.20963183706,
+      "required_pages_per_second_upper_estimate": 0.37198355115810544
     }
   ],
   "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": 0.37947091325000315,
-    "producer_p95_seconds": 0.22453544400195824,
-    "gpu_update_seconds": 0.9531513857520789,
-    "producer_batches_per_second": 2.6352480917059897,
-    "gpu_batches_per_second": 1.049151283781595,
-    "producer_headroom": 2.5117903704100857,
+    "producer_batch_seconds": 0.4869148282421918,
+    "producer_p95_seconds": 0.22407763799856184,
+    "gpu_update_seconds": 0.9533724187168435,
+    "producer_batches_per_second": 2.053747271591818,
+    "gpu_batches_per_second": 1.0489080451329955,
+    "producer_headroom": 1.9579860037506096,
     "sustainable": true,
-    "buffer_seconds": 60.04853730238097,
+    "buffer_seconds": 60.06246237916114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347359.11500526336,
-    "upload_bytes_per_second": 4764600.387720583,
+    "generation_bytes_per_second": 301798.2087931432,
+    "upload_bytes_per_second": 4776004.706385802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.53517997377801,
+    "estimated_full_shards_per_hour": 64.7588701042721,
     "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": 1.2277038499996706,
-      "bytes": 10100946,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9283436340010667,
       "bytes": 2568999,
@@ -562,6 +556,12 @@
       "bytes": 6745128,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3833632300011232,
+      "bytes": 11874470,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 944200, "loss": 0.49090301990509033, "lr": 3e-05, "grad_norm": 0.33144766817361304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138204714657, "tokens_per_second": 152904.86646917247, "data_wait_s": 8.692000847076997e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 944300, "loss": 0.48684871196746826, "lr": 3e-05, "grad_norm": 0.33078584826771573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138219204026, "tokens_per_second": 150047.56068361085, "data_wait_s": 1.2052998499711975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944400, "loss": 0.5279709100723267, "lr": 3e-05, "grad_norm": 0.32609924167972903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138233576903, "tokens_per_second": 144423.03887124828, "data_wait_s": 1.3274002412799746e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 944500, "loss": 0.577720582485199, "lr": 3e-05, "grad_norm": 0.35585286234893304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138248100173, "tokens_per_second": 150613.3662130495, "data_wait_s": 1.7441001546103507e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44b24a56bff85d1b55071f0a4f58e78199b976bddb4d990b0bd9155f9c486f49
-size 413406
+oid sha256:dccfee78f62bd553455e586d7536ba2c3b8dfaea81d6b34cc845f2f0a9c7ecf2
+size 415524