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": 76722.65394123836,
-      "required_pages_per_second_upper_estimate": 0.005307851220738532
+      "required_token_positions_per_second": 75686.7385809766,
+      "required_pages_per_second_upper_estimate": 0.005236184192460829
     },
     {
       "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": 9590.331742654795,
-      "required_pages_per_second_upper_estimate": 0.00401527163174944
+      "required_token_positions_per_second": 9460.842322622075,
+      "required_pages_per_second_upper_estimate": 0.003961057115628318
     },
     {
       "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": 19180.66348530959,
-      "required_pages_per_second_upper_estimate": 0.0039948263758784125
+      "required_token_positions_per_second": 18921.68464524415,
+      "required_pages_per_second_upper_estimate": 0.003940887913224073
     },
     {
       "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": 28770.995227964384,
-      "required_pages_per_second_upper_estimate": 0.0017247772331504257
+      "required_token_positions_per_second": 28382.526967866226,
+      "required_pages_per_second_upper_estimate": 0.0017014891541142292
     },
     {
       "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": 28770.995227964384,
-      "required_pages_per_second_upper_estimate": 0.002236818328740464
+      "required_token_positions_per_second": 28382.526967866226,
+      "required_pages_per_second_upper_estimate": 0.002206616630209128
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28770.995227964384,
-      "required_pages_per_second_upper_estimate": 0.3690434348964788
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28382.526967866226,
+      "required_pages_per_second_upper_estimate": 0.36406058116066015
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13286256108625594,
-    "producer_p95_seconds": 0.2986777659971267,
-    "gpu_update_seconds": 0.9609678004161333,
-    "producer_batches_per_second": 7.526574768875546,
-    "gpu_batches_per_second": 1.0406175936040358,
-    "producer_headroom": 7.232796000313901,
+    "producer_batch_seconds": 0.08630933592195333,
+    "producer_p95_seconds": 0.3049622050020844,
+    "gpu_update_seconds": 0.974120452040869,
+    "producer_batches_per_second": 11.586232118669836,
+    "gpu_batches_per_second": 1.0265670922984023,
+    "producer_headroom": 11.286385668889096,
     "sustainable": true,
-    "buffer_seconds": 60.540971426216394,
+    "buffer_seconds": 60.39546802653388,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 42012.24454418529,
-    "upload_bytes_per_second": 5274330.074873835,
+    "generation_bytes_per_second": 39497.36858662972,
+    "upload_bytes_per_second": 5333695.414676675,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 9.014849684182826,
+    "estimated_full_shards_per_hour": 8.475215847007453,
     "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": 1.2405726069991942,
-      "bytes": 6889170,
-      "pages": 43,
-      "shards": 2
-    },
     {
       "seconds": 0.9624333549982111,
       "bytes": 484343,
@@ -562,6 +556,12 @@
       "bytes": 789312,
       "pages": 22,
       "shards": 1
+    },
+    {
+      "seconds": 1.4157909589994233,
+      "bytes": 12468598,
+      "pages": 49,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 718500, "loss": 0.5605199337005615, "lr": 8e-05, "grad_norm": 0.23433685480020014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105168631666, "tokens_per_second": 153220.88380636225, "data_wait_s": 1.1659998563118279e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718600, "loss": 0.5746077299118042, "lr": 8e-05, "grad_norm": 0.22633490972812828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105183270447, "tokens_per_second": 136851.05222668315, "data_wait_s": 1.6204998246394098e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718700, "loss": 0.5621911287307739, "lr": 8e-05, "grad_norm": 0.22148633925331962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105197791540, "tokens_per_second": 151198.74785465543, "data_wait_s": 9.552997653372586e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 718800, "loss": 0.5639420747756958, "lr": 8e-05, "grad_norm": 0.21778007256652135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105212535088, "tokens_per_second": 154353.76146714773, "data_wait_s": 1.52729990077205e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ba036362b5b8908bfd8af561240d6794b791b4742ce66a054728aa8c7a62de5
-size 718629
+oid sha256:571d548786f158fec90dc47cbcbe5d5bf2aef30a30c6f1dbc4c3a064657b5ceb
+size 720747