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": 74786.18258862563,
-      "required_pages_per_second_upper_estimate": 0.006482276762935193
+      "required_token_positions_per_second": 76045.16265563831,
+      "required_pages_per_second_upper_estimate": 0.006591401964288045
     },
     {
       "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": 9348.272823578203,
-      "required_pages_per_second_upper_estimate": 0.0037291657984594717
+      "required_token_positions_per_second": 9505.645331954789,
+      "required_pages_per_second_upper_estimate": 0.003791944044979571
     },
     {
       "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": 18696.545647156407,
-      "required_pages_per_second_upper_estimate": 0.004325116787504628
+      "required_token_positions_per_second": 19011.290663909578,
+      "required_pages_per_second_upper_estimate": 0.004397927507807371
     },
     {
       "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": 28044.81847073461,
-      "required_pages_per_second_upper_estimate": 0.001707782305076784
+      "required_token_positions_per_second": 28516.93599586437,
+      "required_pages_per_second_upper_estimate": 0.0017365317853479654
     },
     {
       "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": 28044.81847073461,
-      "required_pages_per_second_upper_estimate": 0.0021589677589034123
+      "required_token_positions_per_second": 28516.93599586437,
+      "required_pages_per_second_upper_estimate": 0.002195312672892856
     },
     {
       "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": 28044.81847073461,
-      "required_pages_per_second_upper_estimate": 0.3482227854369372
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28516.93599586437,
+      "required_pages_per_second_upper_estimate": 0.35408490502333545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.11449109225787879,
-    "producer_p95_seconds": 0.30624731199714006,
-    "gpu_update_seconds": 0.9858505601971109,
-    "producer_batches_per_second": 8.734303955696468,
-    "gpu_batches_per_second": 1.01435251991951,
-    "producer_headroom": 8.610718447655206,
+    "producer_batch_seconds": 0.08408742583588946,
+    "producer_p95_seconds": 0.2333088890009094,
+    "gpu_update_seconds": 0.9695291248684507,
+    "producer_batches_per_second": 11.892384504095364,
+    "gpu_batches_per_second": 1.031428529943011,
+    "producer_headroom": 11.530013140854702,
     "sustainable": true,
-    "buffer_seconds": 60.13688417202377,
+    "buffer_seconds": 60.11080574184394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335670.20117900206,
+    "generation_bytes_per_second": 322567.7788262318,
     "upload_bytes_per_second": 5860611.2706689155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.02701117064997,
+    "estimated_full_shards_per_hour": 69.21553634252753,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -176,3 +176,4 @@
 {"step": 132600, "loss": 0.6222973987460136, "lr": 0.0003, "grad_norm": 0.14609830944624297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19405003107, "tokens_per_second": 149007.31623234323, "data_wait_s": 1.2625001545529813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132700, "loss": 0.595722945407033, "lr": 0.0003, "grad_norm": 0.16597022553613952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19419602553, "tokens_per_second": 148046.76957254668, "data_wait_s": 1.2183998478576541e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132800, "loss": 0.6088363602757454, "lr": 0.0003, "grad_norm": 0.13583344653923674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19434292909, "tokens_per_second": 150707.80683408817, "data_wait_s": 1.588300074217841e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 132900, "loss": 0.5888306833803654, "lr": 0.0003, "grad_norm": 0.12400495761659745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19448992779, "tokens_per_second": 149753.48412678338, "data_wait_s": 1.1275998986093327e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd8870f9c200a8c4871a75d7c181b0dc1aaffb80591dd441687b499e7758fbf0
-size 379518
+oid sha256:428d88ab72aef0157db6bbc98218658e4fc0df1382db161961c95203501150a2
+size 381636