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": 73439.19842918367,
-      "required_pages_per_second_upper_estimate": 0.006365523589894904
+      "required_token_positions_per_second": 73753.741159052,
+      "required_pages_per_second_upper_estimate": 0.006392787356518631
     },
     {
       "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": 9179.899803647959,
-      "required_pages_per_second_upper_estimate": 0.003661999283408313
+      "required_token_positions_per_second": 9219.2176448815,
+      "required_pages_per_second_upper_estimate": 0.003677683758130485
     },
     {
       "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": 18359.799607295918,
-      "required_pages_per_second_upper_estimate": 0.00424721651771079
+      "required_token_positions_per_second": 18438.435289763,
+      "required_pages_per_second_upper_estimate": 0.0042654074989088055
     },
     {
       "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": 27539.69941094388,
-      "required_pages_per_second_upper_estimate": 0.0016770232044904204
+      "required_token_positions_per_second": 27657.652934644502,
+      "required_pages_per_second_upper_estimate": 0.0016842059552294771
     },
     {
       "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": 27539.69941094388,
-      "required_pages_per_second_upper_estimate": 0.0021200822954216716
+      "required_token_positions_per_second": 27657.652934644502,
+      "required_pages_per_second_upper_estimate": 0.002129162684192397
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27539.69941094388,
-      "required_pages_per_second_upper_estimate": 0.3419508972390811
+      "required_token_positions_per_second": 27657.652934644502,
+      "required_pages_per_second_upper_estimate": 0.34341548523839355
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11654920008584213,
-    "producer_p95_seconds": 0.4383445470011793,
-    "gpu_update_seconds": 1.0039325261848386,
-    "producer_batches_per_second": 8.580067467331125,
-    "gpu_batches_per_second": 0.9960828779999956,
-    "producer_headroom": 8.613808807314086,
+    "producer_batch_seconds": 0.142378784015591,
+    "producer_p95_seconds": 0.46177781200094614,
+    "gpu_update_seconds": 0.9996509850395725,
+    "producer_batches_per_second": 7.0235183346593,
+    "gpu_batches_per_second": 1.0003491368143989,
+    "producer_headroom": 7.021067021685668,
     "sustainable": true,
-    "buffer_seconds": 60.23595157109032,
+    "buffer_seconds": 60.97871008741392,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379466.9310410222,
-    "upload_bytes_per_second": 4376941.838641347,
+    "generation_bytes_per_second": 459165.9742724774,
+    "upload_bytes_per_second": 4414925.624233693,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.424769863348,
+    "estimated_full_shards_per_hour": 98.52632924204579,
     "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": 0.9765169189995504,
-      "bytes": 821604,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1198572450011852,
       "bytes": 7695575,
@@ -562,6 +556,12 @@
       "bytes": 7649365,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.291299133998109,
+      "bytes": 5845267,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 278700, "loss": 0.5751864071935415, "lr": 0.0003, "grad_norm": 0.15900936643717134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40794039350, "tokens_per_second": 143447.7581870635, "data_wait_s": 1.870400592451915e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278800, "loss": 0.6008367203176022, "lr": 0.0003, "grad_norm": 0.16020589139534827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40808645024, "tokens_per_second": 145526.61215152068, "data_wait_s": 1.867199898697436e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278900, "loss": 0.5624541845172644, "lr": 0.0003, "grad_norm": 0.1875007228420559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40823246291, "tokens_per_second": 145586.3860545625, "data_wait_s": 1.8457998521625996e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 279000, "loss": 0.6325334496796131, "lr": 0.0003, "grad_norm": 0.1718096946910704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40838042164, "tokens_per_second": 146793.23999873843, "data_wait_s": 2.1006999304518104e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:268a0047eeefdfca33a2c6b4a41aaa3a41f90fbc76e7907f0d80ce691aad60f7
-size 891753
+oid sha256:6e041881cf127c27ada9b8ee2eb222d2144f1eee0330db6407531d16c2ad8a3c
+size 893871