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": 74696.54163580574,
-      "required_pages_per_second_upper_estimate": 0.0064745069123376825
+      "required_token_positions_per_second": 75839.54600321392,
+      "required_pages_per_second_upper_estimate": 0.00657357963398654
     },
     {
       "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": 9337.067704475718,
-      "required_pages_per_second_upper_estimate": 0.00372469590891803
+      "required_token_positions_per_second": 9479.94325040174,
+      "required_pages_per_second_upper_estimate": 0.003781691100367696
     },
     {
       "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": 18674.135408951435,
-      "required_pages_per_second_upper_estimate": 0.00431993257330263
+      "required_token_positions_per_second": 18959.88650080348,
+      "required_pages_per_second_upper_estimate": 0.004386036059355147
     },
     {
       "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": 28011.203113427153,
-      "required_pages_per_second_upper_estimate": 0.0017057353061829094
+      "required_token_positions_per_second": 28439.82975120522,
+      "required_pages_per_second_upper_estimate": 0.001731836419593424
     },
     {
       "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": 28011.203113427153,
-      "required_pages_per_second_upper_estimate": 0.0021563799556446194
+      "required_token_positions_per_second": 28439.82975120522,
+      "required_pages_per_second_upper_estimate": 0.0021893768207352487
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28011.203113427153,
-      "required_pages_per_second_upper_estimate": 0.34780539520254233
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28439.82975120522,
+      "required_pages_per_second_upper_estimate": 0.3531275035226693
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4935290651094988,
-    "producer_p95_seconds": 0.4198398630032898,
-    "gpu_update_seconds": 0.9870336482172359,
-    "producer_batches_per_second": 2.026223115710786,
-    "gpu_batches_per_second": 1.0131366866835632,
-    "producer_headroom": 1.9999503940021117,
+    "producer_batch_seconds": 0.12315990569510404,
+    "producer_p95_seconds": 0.4178811950041563,
+    "gpu_update_seconds": 0.9721577183080126,
+    "producer_batches_per_second": 8.119525541661346,
+    "gpu_batches_per_second": 1.0286396756078278,
+    "producer_headroom": 7.893459424325124,
     "sustainable": true,
-    "buffer_seconds": 60.209052541251395,
+    "buffer_seconds": 60.27377853509678,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322817.8489874752,
-    "upload_bytes_per_second": 3608499.282555691,
+    "generation_bytes_per_second": 322764.41944456607,
+    "upload_bytes_per_second": 3641902.377202831,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.26919557779496,
+    "estimated_full_shards_per_hour": 69.25773084166275,
     "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": 0.9623997290036641,
-      "bytes": 6092433,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.7187483379966579,
       "bytes": 1132933,
@@ -562,6 +556,12 @@
       "bytes": 628978,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 1.3523500509982114,
+      "bytes": 11328528,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 40700, "loss": 0.6462833769619465, "lr": 0.0003, "grad_norm": 0.12658802811712694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5955780384, "tokens_per_second": 149595.9284136709, "data_wait_s": 1.1096999514847994e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40800, "loss": 0.6299373246729374, "lr": 0.0003, "grad_norm": 0.127283173593747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5970591672, "tokens_per_second": 150137.56871674163, "data_wait_s": 1.2154996511526406e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40900, "loss": 0.656970776617527, "lr": 0.0003, "grad_norm": 0.14349440390525414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5985111871, "tokens_per_second": 129470.72505672603, "data_wait_s": 1.4625002222601324e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 41000, "loss": 0.633791621774435, "lr": 0.0003, "grad_norm": 0.13616520934006598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5999586229, "tokens_per_second": 148245.87926764408, "data_wait_s": 9.536001016385853e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b83657b62607479b7a6445f0012d4d6bc161608cd18f34ff1ca92f95ca7737d5
-size 866917
+oid sha256:1347835d5300c260d66bf2d72778bb22a7cee8a2288940341b6b20bc38f00305
+size 869035