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": 77614.68086113984,
-      "required_pages_per_second_upper_estimate": 0.005369563712844932
+      "required_token_positions_per_second": 77582.9583954502,
+      "required_pages_per_second_upper_estimate": 0.005367369079062261
     },
     {
       "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": 9701.83510764248,
-      "required_pages_per_second_upper_estimate": 0.004061955762214746
+      "required_token_positions_per_second": 9697.869799431275,
+      "required_pages_per_second_upper_estimate": 0.0040602955704717655
     },
     {
       "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": 19403.67021528496,
-      "required_pages_per_second_upper_estimate": 0.0040412727966493275
+      "required_token_positions_per_second": 19395.73959886255,
+      "required_pages_per_second_upper_estimate": 0.004039621058392959
     },
     {
       "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": 29105.505322927445,
-      "required_pages_per_second_upper_estimate": 0.0017448306025761205
+      "required_token_positions_per_second": 29093.609398293826,
+      "required_pages_per_second_upper_estimate": 0.0017441174600583602
     },
     {
       "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": 29105.505322927445,
-      "required_pages_per_second_upper_estimate": 0.002262825017269432
+      "required_token_positions_per_second": 29093.609398293826,
+      "required_pages_per_second_upper_estimate": 0.0022619001614538107
     },
     {
       "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": 29105.505322927445,
-      "required_pages_per_second_upper_estimate": 0.37333417122570833
+      "required_token_positions_per_second": 29093.609398293826,
+      "required_pages_per_second_upper_estimate": 0.37318158307735694
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08289884385933988,
-    "producer_p95_seconds": 0.2110502430004999,
-    "gpu_update_seconds": 0.9499233802417677,
-    "producer_batches_per_second": 12.062894407752758,
-    "gpu_batches_per_second": 1.0527164830341234,
-    "producer_headroom": 11.458825431312016,
+    "producer_batch_seconds": 0.060309696515616906,
+    "producer_p95_seconds": 0.2083215860002383,
+    "gpu_update_seconds": 0.9503117891457427,
+    "producer_batches_per_second": 16.58108161331992,
+    "gpu_batches_per_second": 1.052286219556345,
+    "producer_headroom": 15.757197333925632,
     "sustainable": true,
-    "buffer_seconds": 60.79509633547313,
+    "buffer_seconds": 60.81995450532753,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367683.09016950027,
-    "upload_bytes_per_second": 5858504.813481178,
+    "generation_bytes_per_second": 368006.0488628882,
+    "upload_bytes_per_second": 5903279.337436248,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.89623192609554,
+    "estimated_full_shards_per_hour": 78.96553134360299,
     "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": 0.9403381189995343,
-      "bytes": 1076403,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.025973077999879,
       "bytes": 6061719,
@@ -562,6 +556,12 @@
       "bytes": 8230638,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2417163159989286,
+      "bytes": 6099161,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -119,3 +119,4 @@
 {"step": 496900, "loss": 0.5843043327331543, "lr": 0.0003, "grad_norm": 0.18514039744330238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72729603112, "tokens_per_second": 153090.59701669193, "data_wait_s": 1.0415000360808335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497000, "loss": 0.5307057499885559, "lr": 0.0003, "grad_norm": 0.15472175292565252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72744074987, "tokens_per_second": 151608.28812211673, "data_wait_s": 1.4495000868919306e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497100, "loss": 0.5368908643722534, "lr": 0.0003, "grad_norm": 0.15766579252660712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72758524455, "tokens_per_second": 151082.53089987973, "data_wait_s": 1.116900057240855e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 497200, "loss": 0.5461172461509705, "lr": 0.0003, "grad_norm": 0.1824614696415162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72773223187, "tokens_per_second": 153278.01444052206, "data_wait_s": 1.3601000318885781e-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 sha256:56cf7612c1c71db2ea2735d3a006ed40c70e1ec5c93926edaec2a304ee8cd93f
-size 258009
+oid sha256:69a6d5834473db105c4086824ba766ccb812cd041bb41751f97230baf9c0e361
+size 260127