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": 75414.2534814217,
-      "required_pages_per_second_upper_estimate": 0.0052173330410210834
+      "required_token_positions_per_second": 77073.77322843882,
+      "required_pages_per_second_upper_estimate": 0.005332142467736051
     },
     {
       "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": 9426.781685177713,
-      "required_pages_per_second_upper_estimate": 0.003946796637997354
+      "required_token_positions_per_second": 9634.221653554852,
+      "required_pages_per_second_upper_estimate": 0.004033647420917733
     },
     {
       "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": 18853.563370355427,
-      "required_pages_per_second_upper_estimate": 0.003926700048143579
+      "required_token_positions_per_second": 19268.443307109705,
+      "required_pages_per_second_upper_estimate": 0.004013108597849805
     },
     {
       "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": 28280.34505553314,
-      "required_pages_per_second_upper_estimate": 0.0016953635044926068
+      "required_token_positions_per_second": 28902.664960664555,
+      "required_pages_per_second_upper_estimate": 0.00173267063257776
     },
     {
       "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": 28280.34505553314,
-      "required_pages_per_second_upper_estimate": 0.0021986724359759637
+      "required_token_positions_per_second": 28902.664960664555,
+      "required_pages_per_second_upper_estimate": 0.002247055071303955
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28280.34505553314,
-      "required_pages_per_second_upper_estimate": 0.3627499013036408
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28902.664960664555,
+      "required_pages_per_second_upper_estimate": 0.37073235285161243
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12065899533587299,
-    "producer_p95_seconds": 0.2596936049994838,
-    "gpu_update_seconds": 0.9776401223432237,
-    "producer_batches_per_second": 8.28781971220915,
-    "gpu_batches_per_second": 1.0228712766034844,
-    "producer_headroom": 8.102505077402734,
+    "producer_batch_seconds": 0.34555559817965786,
+    "producer_p95_seconds": 0.29172671400010586,
+    "gpu_update_seconds": 0.9565899904949212,
+    "producer_batches_per_second": 2.8938903182812563,
+    "gpu_batches_per_second": 1.04537995372774,
+    "producer_headroom": 2.7682665120580117,
     "sustainable": true,
-    "buffer_seconds": 60.61368758527987,
+    "buffer_seconds": 60.26516940118004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376405.7676728683,
-    "upload_bytes_per_second": 4824617.76304396,
+    "generation_bytes_per_second": 346946.66195035336,
+    "upload_bytes_per_second": 4847038.369000407,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.7679154647783,
+    "estimated_full_shards_per_hour": 74.44667714961005,
     "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": 0.9880936410008871,
-      "bytes": 2162409,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.2899976480002806,
       "bytes": 11585751,
@@ -562,6 +556,12 @@
       "bytes": 8285983,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.3943405380014156,
+      "bytes": 6107726,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 786600, "loss": 0.5553094744682312, "lr": 3e-05, "grad_norm": 0.28255089079386364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115135798880, "tokens_per_second": 150367.19231246604, "data_wait_s": 9.563998901285231e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 786700, "loss": 0.5412750840187073, "lr": 3e-05, "grad_norm": 0.27167565713649355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115150412062, "tokens_per_second": 151988.70076290454, "data_wait_s": 1.2495002010837197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786800, "loss": 0.48189279437065125, "lr": 3e-05, "grad_norm": 0.24508727685326637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115165046750, "tokens_per_second": 151457.01707543465, "data_wait_s": 1.237800097442232e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 786900, "loss": 0.5313212871551514, "lr": 3e-05, "grad_norm": 0.25772295429552444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115179794041, "tokens_per_second": 153177.10567686593, "data_wait_s": 1.2923999747727066e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3587a733ceff763dcfaf489399c19dbc3ae4877fddf0df2ce46c1a0e65fe5eee
-size 571704
+oid sha256:cb49f6260a71641686650958b8161913245c6921ab5e3f07ec61bfbb7428644e
+size 573822