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": 76321.27758658606,
-      "required_pages_per_second_upper_estimate": 0.005280083073202232
+      "required_token_positions_per_second": 76235.28940132579,
+      "required_pages_per_second_upper_estimate": 0.0052741342110258485
     },
     {
       "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": 9540.159698323258,
-      "required_pages_per_second_upper_estimate": 0.00399426564449925
+      "required_token_positions_per_second": 9529.411175165724,
+      "required_pages_per_second_upper_estimate": 0.003989765462307878
     },
     {
       "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": 19080.319396646515,
-      "required_pages_per_second_upper_estimate": 0.003973927348461465
+      "required_token_positions_per_second": 19058.822350331448,
+      "required_pages_per_second_upper_estimate": 0.003969450080629271
     },
     {
       "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": 28620.47909496977,
-      "required_pages_per_second_upper_estimate": 0.0017157540208022264
+      "required_token_positions_per_second": 28588.23352549717,
+      "required_pages_per_second_upper_estimate": 0.0017138209481484775
     },
     {
       "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": 28620.47909496977,
-      "required_pages_per_second_upper_estimate": 0.0022251163614506343
+      "required_token_positions_per_second": 28588.23352549717,
+      "required_pages_per_second_upper_estimate": 0.00222260941025741
     },
     {
       "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": 28620.47909496977,
-      "required_pages_per_second_upper_estimate": 0.36711277555405614
+      "required_token_positions_per_second": 28588.23352549717,
+      "required_pages_per_second_upper_estimate": 0.36669916401145664
     }
   ],
   "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": 0.08146680923445615,
-    "producer_p95_seconds": 0.21921908900185372,
-    "gpu_update_seconds": 0.9660215647773454,
-    "producer_batches_per_second": 12.274937602159739,
-    "gpu_batches_per_second": 1.0351735783770895,
-    "producer_headroom": 11.857854429982625,
+    "producer_batch_seconds": 0.06858718393758068,
+    "producer_p95_seconds": 0.22669146000043838,
+    "gpu_update_seconds": 0.9671111709417584,
+    "producer_batches_per_second": 14.579983352430284,
+    "gpu_batches_per_second": 1.0340072889719751,
+    "producer_headroom": 14.100464772280196,
     "sustainable": true,
-    "buffer_seconds": 60.85935858097276,
+    "buffer_seconds": 60.92800376933078,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 388695.5589407748,
-    "upload_bytes_per_second": 3470949.65222427,
+    "generation_bytes_per_second": 389341.2469068264,
+    "upload_bytes_per_second": 3447999.215230209,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.40501857917245,
+    "estimated_full_shards_per_hour": 83.54356818584054,
     "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": 1.8303398019997985,
-      "bytes": 12284688,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.182235641999796,
       "bytes": 6603515,
@@ -562,6 +556,12 @@
       "bytes": 2445888,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.9226194889997714,
+      "bytes": 9744014,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -172,3 +172,4 @@
 {"step": 859700, "loss": 0.5077232718467712, "lr": 3e-05, "grad_norm": 0.29974919063921357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125836052334, "tokens_per_second": 153310.24996200306, "data_wait_s": 1.0931002179859206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859800, "loss": 0.5487866997718811, "lr": 3e-05, "grad_norm": 0.299327830126915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125850669999, "tokens_per_second": 151724.1533317882, "data_wait_s": 1.35770023916848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859900, "loss": 0.5440846085548401, "lr": 3e-05, "grad_norm": 0.3083464456228531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125865178178, "tokens_per_second": 150252.7886356925, "data_wait_s": 8.582999726058915e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 860000, "loss": 0.5167421102523804, "lr": 3e-05, "grad_norm": 0.3017758321480747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125879763724, "tokens_per_second": 150859.77013761847, "data_wait_s": 1.5797999367350712e-05, "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:f3cde9c6377f1f484146c7a2143fba313f3f9cf3394090eea8f6cf9d3fc48207
-size 371046
+oid sha256:dcd73d543487c4c035a0a5470c18d6f79edb7926d59c0309c255113b591f1266
+size 373164