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": 75709.40404438647,
-      "required_pages_per_second_upper_estimate": 0.0052377522418106445
+      "required_token_positions_per_second": 76198.91242621694,
+      "required_pages_per_second_upper_estimate": 0.005271617567481607
     },
     {
       "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": 9463.675505548308,
-      "required_pages_per_second_upper_estimate": 0.003962243310155945
+      "required_token_positions_per_second": 9524.864053277117,
+      "required_pages_per_second_upper_estimate": 0.003987861677327821
     },
     {
       "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": 18927.351011096616,
-      "required_pages_per_second_upper_estimate": 0.003942068067799026
+      "required_token_positions_per_second": 19049.728106554234,
+      "required_pages_per_second_upper_estimate": 0.0039675559894818135
     },
     {
       "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": 28391.026516644928,
-      "required_pages_per_second_upper_estimate": 0.001701998689085453
+      "required_token_positions_per_second": 28574.59215983135,
+      "required_pages_per_second_upper_estimate": 0.001713003169105971
     },
     {
       "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": 28391.026516644928,
-      "required_pages_per_second_upper_estimate": 0.0022072774327411082
+      "required_token_positions_per_second": 28574.59215983135,
+      "required_pages_per_second_upper_estimate": 0.002221548854078919
     },
     {
       "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": 28391.026516644928,
-      "required_pages_per_second_upper_estimate": 0.3641696042462889
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28574.59215983135,
+      "required_pages_per_second_upper_estimate": 0.3665241872196528
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5145181251953517,
-    "producer_p95_seconds": 0.22991459600052622,
-    "gpu_update_seconds": 0.9738288252378156,
-    "producer_batches_per_second": 1.9435661272774813,
-    "gpu_batches_per_second": 1.02687451232078,
-    "producer_headroom": 1.8927007184986404,
+    "producer_batch_seconds": 0.07912105367975641,
+    "producer_p95_seconds": 0.19795135000094888,
+    "gpu_update_seconds": 0.9675728649196994,
+    "producer_batches_per_second": 12.638861004651355,
+    "gpu_batches_per_second": 1.033513894669826,
+    "producer_headroom": 12.229018951592382,
     "sustainable": true,
-    "buffer_seconds": 60.37738716474456,
+    "buffer_seconds": 60.95709048994106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322114.11835021316,
-    "upload_bytes_per_second": 4035311.315869877,
+    "generation_bytes_per_second": 319667.2788768152,
+    "upload_bytes_per_second": 4044348.2593192942,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.11819136504933,
+    "estimated_full_shards_per_hour": 68.59315657356589,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 10476032,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.5808353909997095,
+      "bytes": 7239609,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 810900, "loss": 0.5189882516860962, "lr": 3e-05, "grad_norm": 0.2833655514270679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118692713724, "tokens_per_second": 150418.51877316384, "data_wait_s": 1.1419000657042488e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811000, "loss": 0.5326195955276489, "lr": 3e-05, "grad_norm": 0.29499657528183704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118707293791, "tokens_per_second": 122921.03879083165, "data_wait_s": 1.7456999557907693e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811100, "loss": 0.48350292444229126, "lr": 3e-05, "grad_norm": 0.2596258116368013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118721958994, "tokens_per_second": 151118.6890138273, "data_wait_s": 1.3261998901725747e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 811200, "loss": 0.5469661355018616, "lr": 3e-05, "grad_norm": 0.29625469743622646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118736633880, "tokens_per_second": 152153.00064827313, "data_wait_s": 1.4392000593943521e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:233b49e75220d88955cf9dc9ba5bf85a997596a946d1f61f3a96c90e20f24f1c
-size 236829
+oid sha256:c33784bf7a4ec4e0669f23fdad8ae28428d588ffe46a01aee21e10cac96bd6e0
+size 238947