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": 76699.89328915931,
-      "required_pages_per_second_upper_estimate": 0.005306276586015715
+      "required_token_positions_per_second": 77033.01004300044,
+      "required_pages_per_second_upper_estimate": 0.005329322375983803
     },
     {
       "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": 9587.486661144914,
-      "required_pages_per_second_upper_estimate": 0.004014080455533311
+      "required_token_positions_per_second": 9629.126255375055,
+      "required_pages_per_second_upper_estimate": 0.004031514084103865
     },
     {
       "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": 19174.97332228983,
-      "required_pages_per_second_upper_estimate": 0.003993641264981087
+      "required_token_positions_per_second": 19258.25251075011,
+      "required_pages_per_second_upper_estimate": 0.004010986123717474
     },
     {
       "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": 28762.459983434743,
-      "required_pages_per_second_upper_estimate": 0.0017242655582734375
+      "required_token_positions_per_second": 28887.378766125163,
+      "required_pages_per_second_upper_estimate": 0.001731754248555791
     },
     {
       "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": 28762.459983434743,
-      "required_pages_per_second_upper_estimate": 0.0022361547510208577
+      "required_token_positions_per_second": 28887.378766125163,
+      "required_pages_per_second_upper_estimate": 0.002245866636915381
     },
     {
       "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": 28762.459983434743,
-      "required_pages_per_second_upper_estimate": 0.3689339539440841
+      "required_token_positions_per_second": 28887.378766125163,
+      "required_pages_per_second_upper_estimate": 0.370536277961098
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07042869415626285,
-    "producer_p95_seconds": 0.17390763699950185,
-    "gpu_update_seconds": 0.9612529670940317,
-    "producer_batches_per_second": 14.198758218933628,
-    "gpu_batches_per_second": 1.0403088825027034,
-    "producer_headroom": 13.64859846700072,
+    "producer_batch_seconds": 0.19538759585155674,
+    "producer_p95_seconds": 0.17317094299960445,
+    "gpu_update_seconds": 0.957096184594688,
+    "producer_batches_per_second": 5.11803216392374,
+    "gpu_batches_per_second": 1.0448270676405225,
+    "producer_headroom": 4.898449056724306,
     "sustainable": true,
-    "buffer_seconds": 60.558936926924,
+    "buffer_seconds": 60.297059629465345,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532750.4510059691,
-    "upload_bytes_per_second": 4149470.1147120805,
+    "generation_bytes_per_second": 536994.1589573388,
+    "upload_bytes_per_second": 4163151.148946668,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.31584499010376,
+    "estimated_full_shards_per_hour": 115.22644592800258,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 1136147,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.611343528999896,
+      "bytes": 7401953,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 846500, "loss": 0.5031244158744812, "lr": 3e-05, "grad_norm": 0.28552692825837883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123903826351, "tokens_per_second": 150534.72185948706, "data_wait_s": 1.5296000128728338e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846600, "loss": 0.5222848653793335, "lr": 3e-05, "grad_norm": 0.28802581083794127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123918695651, "tokens_per_second": 154192.15765812015, "data_wait_s": 1.5055000403663144e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846700, "loss": 0.4852267801761627, "lr": 3e-05, "grad_norm": 0.2792860797559969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123933412034, "tokens_per_second": 152542.14170023164, "data_wait_s": 2.0462000065890606e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 846800, "loss": 0.5162094831466675, "lr": 3e-05, "grad_norm": 0.29515991358143606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123948088795, "tokens_per_second": 153339.32341996324, "data_wait_s": 8.813000022200868e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f10608740bbd0c1a4a1007368dcd0a6f5f5298d549bc50548afc11b29590e390
-size 89904
+oid sha256:a1a798a78c44652a45e33729029f73ecd133291b5ff03458dfd293c9f12bddea
+size 92022