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": 76062.27200574274,
-      "required_pages_per_second_upper_estimate": 0.005262164466143746
+      "required_token_positions_per_second": 76365.90682755141,
+      "required_pages_per_second_upper_estimate": 0.005283170627646321
     },
     {
       "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": 9507.784000717842,
-      "required_pages_per_second_upper_estimate": 0.003980710615993309
+      "required_token_positions_per_second": 9545.738353443927,
+      "required_pages_per_second_upper_estimate": 0.003996601310902709
     },
     {
       "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": 19015.568001435684,
-      "required_pages_per_second_upper_estimate": 0.0039604413404481726
+      "required_token_positions_per_second": 19091.476706887854,
+      "required_pages_per_second_upper_estimate": 0.0039762511219466785
     },
     {
       "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": 28523.352002153526,
-      "required_pages_per_second_upper_estimate": 0.001709931399892376
+      "required_token_positions_per_second": 28637.215060331782,
+      "required_pages_per_second_upper_estimate": 0.0017167573163713395
     },
     {
       "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": 28523.352002153526,
-      "required_pages_per_second_upper_estimate": 0.0022175651572011027
+      "required_token_positions_per_second": 28637.215060331782,
+      "required_pages_per_second_upper_estimate": 0.002226417509143799
     },
     {
       "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": 28523.352002153526,
-      "required_pages_per_second_upper_estimate": 0.3658669334943565
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28637.215060331782,
+      "required_pages_per_second_upper_estimate": 0.36732744654804045
     }
   ],
   "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": 0.08309368866402167,
-    "producer_p95_seconds": 0.22190800400130684,
-    "gpu_update_seconds": 0.9693110402281108,
-    "producer_batches_per_second": 12.034608356879758,
-    "gpu_batches_per_second": 1.0316605903556686,
-    "producer_headroom": 11.665278745145033,
+    "producer_batch_seconds": 0.08457822993750597,
+    "producer_p95_seconds": 0.23299581900027988,
+    "gpu_update_seconds": 0.9654570090615396,
+    "producer_batches_per_second": 11.823373470204926,
+    "gpu_batches_per_second": 1.0357789011983427,
+    "producer_headroom": 11.414958787561604,
     "sustainable": true,
-    "buffer_seconds": 60.09728449414287,
+    "buffer_seconds": 60.823791570877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408203.8271353245,
-    "upload_bytes_per_second": 4990927.859828072,
+    "generation_bytes_per_second": 411379.8747374496,
+    "upload_bytes_per_second": 4936420.297817216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.59103880448151,
+    "estimated_full_shards_per_hour": 88.27254468529334,
     "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": 1.2741041360004601,
-      "bytes": 11482818,
-      "pages": 21,
-      "shards": 3
-    },
     {
       "seconds": 1.076980086999356,
       "bytes": 1228924,
@@ -562,6 +556,12 @@
       "bytes": 7578967,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.20693444599965,
+      "bytes": 6453330,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 732800, "loss": 0.6318305134773254, "lr": 5e-05, "grad_norm": 0.2544219996035508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107261768006, "tokens_per_second": 150330.6799940051, "data_wait_s": 1.886000063677784e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732900, "loss": 0.5464026927947998, "lr": 5e-05, "grad_norm": 0.24496951254818455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107276460721, "tokens_per_second": 150820.83514368258, "data_wait_s": 2.204400152550079e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 733000, "loss": 0.5088445544242859, "lr": 5e-05, "grad_norm": 0.21719846150426592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107290984184, "tokens_per_second": 149440.16431609707, "data_wait_s": 1.0574000043561682e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 733100, "loss": 0.5551868677139282, "lr": 5e-05, "grad_norm": 0.22673373222473278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107305563008, "tokens_per_second": 149502.00366503536, "data_wait_s": 1.994199919863604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39d052a816a757d32f27fcc1aefe9fbdd0de807cef7bee2e64689aec84def308
-size 277071
+oid sha256:e7ffbee9db582e3bc693d53114528fb4e76458d762450ac153deba2bc404cffe
+size 279189