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": 75335.21015787999,
-      "required_pages_per_second_upper_estimate": 0.0065298650811405085
+      "required_token_positions_per_second": 75649.29357582402,
+      "required_pages_per_second_upper_estimate": 0.006557089035770751
     },
     {
       "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": 9416.901269734999,
-      "required_pages_per_second_upper_estimate": 0.003756542711718126
+      "required_token_positions_per_second": 9456.161696978003,
+      "required_pages_per_second_upper_estimate": 0.0037722042831410574
     },
     {
       "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": 18833.802539469998,
-      "required_pages_per_second_upper_estimate": 0.004356868753902577
+      "required_token_positions_per_second": 18912.323393956005,
+      "required_pages_per_second_upper_estimate": 0.004375033171668074
     },
     {
       "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": 28250.703809204995,
-      "required_pages_per_second_upper_estimate": 0.0017203196419927416
+      "required_token_positions_per_second": 28368.485090934006,
+      "required_pages_per_second_upper_estimate": 0.0017274919040994108
     },
     {
       "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": 28250.703809204995,
-      "required_pages_per_second_upper_estimate": 0.002174817382185956
+      "required_token_positions_per_second": 28368.485090934006,
+      "required_pages_per_second_upper_estimate": 0.0021838845112929104
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28250.703809204995,
-      "required_pages_per_second_upper_estimate": 0.35077919228683707
+      "required_token_positions_per_second": 28368.485090934006,
+      "required_pages_per_second_upper_estimate": 0.3522416416173188
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.33905480934366494,
-    "producer_p95_seconds": 0.2216366989996459,
-    "gpu_update_seconds": 0.9786658833962,
-    "producer_batches_per_second": 2.9493756538530707,
-    "gpu_batches_per_second": 1.0217991829139539,
-    "producer_headroom": 2.8864533297453603,
+    "producer_batch_seconds": 0.09909281449212415,
+    "producer_p95_seconds": 0.2560940559997107,
+    "gpu_update_seconds": 0.9746026236993438,
+    "producer_batches_per_second": 10.091549070689474,
+    "gpu_batches_per_second": 1.0260592119116756,
+    "producer_headroom": 9.835250201484637,
     "sustainable": true,
-    "buffer_seconds": 60.6772847705644,
+    "buffer_seconds": 60.425362669359316,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412045.0716109421,
+    "generation_bytes_per_second": 411530.24745157396,
     "upload_bytes_per_second": 4723942.654699654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.41528044935414,
+    "estimated_full_shards_per_hour": 88.3048111692468,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 252200, "loss": 0.5551466159522533, "lr": 0.0003, "grad_norm": 0.16295606531543627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36915240813, "tokens_per_second": 148722.7453244018, "data_wait_s": 1.4395998732652515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 252300, "loss": 0.5775285307317972, "lr": 0.0003, "grad_norm": 0.15426928125046294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36929798971, "tokens_per_second": 148357.1662204022, "data_wait_s": 1.1378000635886565e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 252400, "loss": 0.6342363152652979, "lr": 0.0003, "grad_norm": 0.15756993999892674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36944491384, "tokens_per_second": 150558.81862218984, "data_wait_s": 1.6976000551949255e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 252500, "loss": 0.6091282796114683, "lr": 0.0003, "grad_norm": 0.1357873190525464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36959122419, "tokens_per_second": 149281.72665783617, "data_wait_s": 1.4854000255581923e-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 sha256:85cd3703e0ecd2b7bebb9ff83ba1eab87b077b850525a95ef6adb5222968e250
-size 326568
+oid sha256:7c18a1080789b4733cb8d6a5e030dcd2c8289679ef37cc6796bdc7ba6c977bba
+size 328686