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": 76609.12409814615,
-      "required_pages_per_second_upper_estimate": 0.005299996962767884
+      "required_token_positions_per_second": 76586.20146199431,
+      "required_pages_per_second_upper_estimate": 0.0052984111216110566
     },
     {
       "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": 9576.140512268268,
-      "required_pages_per_second_upper_estimate": 0.0040093300599331905
+      "required_token_positions_per_second": 9573.275182749288,
+      "required_pages_per_second_upper_estimate": 0.004008130406298478
     },
     {
       "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": 19152.281024536536,
-      "required_pages_per_second_upper_estimate": 0.0039889150577952105
+      "required_token_positions_per_second": 19146.550365498577,
+      "required_pages_per_second_upper_estimate": 0.003987721512645245
     },
     {
       "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": 28728.421536804806,
-      "required_pages_per_second_upper_estimate": 0.0017222250053716714
+      "required_token_positions_per_second": 28719.825548247867,
+      "required_pages_per_second_upper_estimate": 0.0017217096889829973
     },
     {
       "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": 28728.421536804806,
-      "required_pages_per_second_upper_estimate": 0.002233508411514684
+      "required_token_positions_per_second": 28719.825548247867,
+      "required_pages_per_second_upper_estimate": 0.0022328401112141397
     },
     {
       "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": 28728.421536804806,
-      "required_pages_per_second_upper_estimate": 0.3684973452983518
+      "required_token_positions_per_second": 28719.825548247867,
+      "required_pages_per_second_upper_estimate": 0.3683870851867968
     }
   ],
   "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": 0.4134595275313444,
-    "producer_p95_seconds": 0.2322301879994484,
-    "gpu_update_seconds": 0.9623918934975022,
-    "producer_batches_per_second": 2.4186164144547133,
-    "gpu_batches_per_second": 1.0390777465568868,
-    "producer_headroom": 2.327656830751211,
+    "producer_batch_seconds": 0.0970638729765767,
+    "producer_p95_seconds": 0.21750977800002147,
+    "gpu_update_seconds": 0.9626799422424327,
+    "producer_batches_per_second": 10.30249431981061,
+    "gpu_batches_per_second": 1.0387668384059559,
+    "producer_headroom": 9.918004636748268,
     "sustainable": true,
-    "buffer_seconds": 60.63068929034264,
+    "buffer_seconds": 60.64883636127326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277218.8678118559,
-    "upload_bytes_per_second": 4825120.841091904,
+    "generation_bytes_per_second": 277562.5954596295,
+    "upload_bytes_per_second": 4872711.539456161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.48471570746191,
+    "estimated_full_shards_per_hour": 59.558471659104,
     "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": 1.0174859720000313,
-      "bytes": 4930803,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.3428281450014765,
       "bytes": 7443989,
@@ -562,6 +556,12 @@
       "bytes": 6680190,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5331677430003765,
+      "bytes": 11639669,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -156,3 +156,4 @@
 {"step": 735600, "loss": 0.5117169618606567, "lr": 5e-05, "grad_norm": 0.2415996822182658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107671466857, "tokens_per_second": 130425.02266456, "data_wait_s": 9.953000699169934e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 735700, "loss": 0.5571478009223938, "lr": 5e-05, "grad_norm": 0.23093229228948703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107686059365, "tokens_per_second": 151357.19492949295, "data_wait_s": 1.6595999113633297e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735800, "loss": 0.5460484027862549, "lr": 5e-05, "grad_norm": 0.21712870145517582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107700734100, "tokens_per_second": 151343.9120267944, "data_wait_s": 9.00899976841174e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 735900, "loss": 0.5330012440681458, "lr": 5e-05, "grad_norm": 0.2299289590874985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107715457366, "tokens_per_second": 151553.5385937623, "data_wait_s": 1.3471000784193166e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:718d79c60f2f1bee0433dbda340bf5b470d307d718ae1b1177951894d4b9b45a
-size 337158
+oid sha256:3d933ef4755f65d5d449a2bef26efa22a409edb720674345a4d394398579b588
+size 339276