Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75744.40124839635,
-      "required_pages_per_second_upper_estimate": 0.006565332727781128
+      "required_token_positions_per_second": 76082.8103878681,
+      "required_pages_per_second_upper_estimate": 0.006594665174300417
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9468.050156049543,
-      "required_pages_per_second_upper_estimate": 0.0037769467672129978
+      "required_token_positions_per_second": 9510.351298483512,
+      "required_pages_per_second_upper_estimate": 0.003793821325388349
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18936.100312099086,
-      "required_pages_per_second_upper_estimate": 0.004380533543221009
+      "required_token_positions_per_second": 19020.702596967025,
+      "required_pages_per_second_upper_estimate": 0.004400104792875847
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28404.15046814863,
-      "required_pages_per_second_upper_estimate": 0.0017296637384500086
+      "required_token_positions_per_second": 28531.05389545054,
+      "required_pages_per_second_upper_estimate": 0.0017373914913618685
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28404.15046814863,
-      "required_pages_per_second_upper_estimate": 0.0021866301307589754
+      "required_token_positions_per_second": 28531.05389545054,
+      "required_pages_per_second_upper_estimate": 0.002196399507883847
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28404.15046814863,
-      "required_pages_per_second_upper_estimate": 0.3526844862379854
+      "required_token_positions_per_second": 28531.05389545054,
+      "required_pages_per_second_upper_estimate": 0.3542602020866253
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2768711365547176,
-    "producer_p95_seconds": 0.28734239600089495,
-    "gpu_update_seconds": 0.9733788740136218,
-    "producer_batches_per_second": 3.6117885469884343,
-    "gpu_batches_per_second": 1.0273491922796814,
-    "producer_headroom": 3.5156386690428976,
+    "producer_batch_seconds": 0.06262007569523576,
+    "producer_p95_seconds": 0.2119504419970326,
+    "gpu_update_seconds": 0.9690493769109824,
+    "producer_batches_per_second": 15.969319565611476,
+    "gpu_batches_per_second": 1.0319391599917005,
+    "producer_headroom": 15.47505917474816,
     "sustainable": true,
-    "buffer_seconds": 60.34949018884455,
+    "buffer_seconds": 60.08106136848091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455169.72782013845,
-    "upload_bytes_per_second": 5330110.05439119,
+    "generation_bytes_per_second": 455298.2353808314,
+    "upload_bytes_per_second": 5377766.3399886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.66882778123012,
+    "estimated_full_shards_per_hour": 97.69640251225192,
     "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": 1.6818293220003397,
-      "bytes": 5841398,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1767334710002615,
       "bytes": 1036234,
@@ -562,6 +556,12 @@
       "bytes": 4790943,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2871860160012147,
+      "bytes": 7434728,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -107,3 +107,4 @@
 {"step": 290900, "loss": 0.5860250983387232, "lr": 0.0003, "grad_norm": 0.13762307720115055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42579513576, "tokens_per_second": 151078.19702101997, "data_wait_s": 1.336499917670153e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291000, "loss": 0.6449759714305401, "lr": 0.0003, "grad_norm": 0.1641612940546156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42594212353, "tokens_per_second": 150595.54317226744, "data_wait_s": 1.2327996955718845e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291100, "loss": 0.5537864081561565, "lr": 0.0003, "grad_norm": 0.14571001136342737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42608853396, "tokens_per_second": 149374.30264801913, "data_wait_s": 1.1989999620709568e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 291200, "loss": 0.5675464998930693, "lr": 0.0003, "grad_norm": 0.16711411768102727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42623302588, "tokens_per_second": 148623.04284251956, "data_wait_s": 9.75500006461516e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9697d294cf34abec7525a85dcb7e7fd30d5f3c3eea2d63d0aa150bcfba0ff1eb
-size 232593
+oid sha256:270cd78696fac01f736783e487574dd7f9d4ff7d2bc02554ac3ff6848093267d
+size 234711