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": 76352.71942845121,
-      "required_pages_per_second_upper_estimate": 0.0052822582927775875
+      "required_token_positions_per_second": 76888.05633681771,
+      "required_pages_per_second_upper_estimate": 0.005319294142251145
     },
     {
       "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": 9544.089928556401,
-      "required_pages_per_second_upper_estimate": 0.003995911149825328
+      "required_token_positions_per_second": 9611.007042102214,
+      "required_pages_per_second_upper_estimate": 0.004023927947878726
     },
     {
       "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": 19088.179857112802,
-      "required_pages_per_second_upper_estimate": 0.00397556447508231
+      "required_token_positions_per_second": 19222.014084204427,
+      "required_pages_per_second_upper_estimate": 0.004003438615139582
     },
     {
       "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": 28632.2697856692,
-      "required_pages_per_second_upper_estimate": 0.0017164608547063148
+      "required_token_positions_per_second": 28833.02112630664,
+      "required_pages_per_second_upper_estimate": 0.001728495591047981
     },
     {
       "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": 28632.2697856692,
-      "required_pages_per_second_upper_estimate": 0.002226033035790747
+      "required_token_positions_per_second": 28833.02112630664,
+      "required_pages_per_second_upper_estimate": 0.0022416405695134807
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28632.2697856692,
-      "required_pages_per_second_upper_estimate": 0.3672640138744911
+      "required_token_positions_per_second": 28833.02112630664,
+      "required_pages_per_second_upper_estimate": 0.3698390365221924
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04544677353125692,
-    "producer_p95_seconds": 0.1449872889998005,
-    "gpu_update_seconds": 0.965623759728548,
-    "producer_batches_per_second": 22.003762254150562,
-    "gpu_batches_per_second": 1.0356000356506512,
-    "producer_headroom": 21.247355636025976,
+    "producer_batch_seconds": 0.09009418627346122,
+    "producer_p95_seconds": 0.14739000500003385,
+    "gpu_update_seconds": 0.9589005563754311,
+    "producer_batches_per_second": 11.099495332192895,
+    "gpu_batches_per_second": 1.0428610071725493,
+    "producer_headroom": 10.643312249526268,
     "sustainable": true,
-    "buffer_seconds": 60.83429686289852,
+    "buffer_seconds": 60.41073505165216,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311451.5461107728,
-    "upload_bytes_per_second": 4960346.847601386,
+    "generation_bytes_per_second": 343718.63382833306,
+    "upload_bytes_per_second": 5079141.230246116,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.83025157444371,
+    "estimated_full_shards_per_hour": 73.75401745927329,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 1114116,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.037127201000203,
+      "bytes": 7612192,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -19,3 +19,4 @@
 {"step": 399000, "loss": 0.5636134743690491, "lr": 0.0003, "grad_norm": 0.1511337301343004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58402707719, "tokens_per_second": 152662.34691461586, "data_wait_s": 9.202000001096167e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 399100, "loss": 0.548101007938385, "lr": 0.0003, "grad_norm": 0.14365877958321444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58417183848, "tokens_per_second": 150143.4592421514, "data_wait_s": 9.580999630998122e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 399200, "loss": 0.5344240069389343, "lr": 0.0003, "grad_norm": 0.16703074589599587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58431709220, "tokens_per_second": 150203.48550130386, "data_wait_s": 8.62199976836564e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 399300, "loss": 0.5939231514930725, "lr": 0.0003, "grad_norm": 0.13250949760920044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58446294719, "tokens_per_second": 151739.47154631745, "data_wait_s": 1.1953000011999393e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f35d0137c48b68dab5756c5b73b2547c21bfe03af9a6632a0ba24c4b912ed3e
-size 44643
+oid sha256:bca6b8691d891def00fe80f75b280d6dba8e02d0ef5352c444d74e5af4dad84c
+size 46761