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": 77459.04002879017,
-      "required_pages_per_second_upper_estimate": 0.005358796118926496
+      "required_token_positions_per_second": 77480.24261228995,
+      "required_pages_per_second_upper_estimate": 0.005360262962849782
     },
     {
       "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": 9682.380003598772,
-      "required_pages_per_second_upper_estimate": 0.004053810316420416
+      "required_token_positions_per_second": 9685.030326536244,
+      "required_pages_per_second_upper_estimate": 0.004054919951289299
     },
     {
       "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": 19364.760007197543,
-      "required_pages_per_second_upper_estimate": 0.0040331688264359095
+      "required_token_positions_per_second": 19370.060653072487,
+      "required_pages_per_second_upper_estimate": 0.0040342728111842285
     },
     {
       "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": 29047.140010796313,
-      "required_pages_per_second_upper_estimate": 0.0017413316912325307
+      "required_token_positions_per_second": 29055.090979608733,
+      "required_pages_per_second_upper_estimate": 0.0017418083396724088
     },
     {
       "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": 29047.140010796313,
-      "required_pages_per_second_upper_estimate": 0.0022582873709731132
+      "required_token_positions_per_second": 29055.090979608733,
+      "required_pages_per_second_upper_estimate": 0.002258905523825663
     },
     {
       "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": 29047.140010796313,
-      "required_pages_per_second_upper_estimate": 0.3725855236694798
+      "required_token_positions_per_second": 29055.090979608733,
+      "required_pages_per_second_upper_estimate": 0.3726875101603203
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0729544766640764,
-    "producer_p95_seconds": 0.15177336000033392,
-    "gpu_update_seconds": 0.9518320905164406,
-    "producer_batches_per_second": 13.707178033838343,
-    "gpu_batches_per_second": 1.0506054691404916,
-    "producer_headroom": 13.046931923029383,
+    "producer_batch_seconds": 0.09419050953126629,
+    "producer_p95_seconds": 0.14681518799989135,
+    "gpu_update_seconds": 0.9515716202507765,
+    "producer_batches_per_second": 10.616780872897314,
+    "gpu_batches_per_second": 1.0508930475842277,
+    "producer_headroom": 10.10262737707035,
     "sustainable": true,
-    "buffer_seconds": 60.9172537930522,
+    "buffer_seconds": 60.900583696049694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435277.74691195576,
-    "upload_bytes_per_second": 4995414.642986172,
+    "generation_bytes_per_second": 433215.1647857955,
+    "upload_bytes_per_second": 4861485.57141088,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.40047173995022,
+    "estimated_full_shards_per_hour": 92.95788963013075,
     "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": 6080107,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1274267459998555,
+      "bytes": 2324297,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 888400, "loss": 0.5353817343711853, "lr": 3e-05, "grad_norm": 0.3351649442302755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130037293845, "tokens_per_second": 153090.013600368, "data_wait_s": 1.27170001178456e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888500, "loss": 0.537222683429718, "lr": 3e-05, "grad_norm": 0.3170551465203647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130051959022, "tokens_per_second": 153041.954718379, "data_wait_s": 1.0757999916677363e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888600, "loss": 0.5856220722198486, "lr": 3e-05, "grad_norm": 0.3560445498797903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130066688750, "tokens_per_second": 153979.4682561426, "data_wait_s": 1.0291999842593214e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 888700, "loss": 0.5092302560806274, "lr": 3e-05, "grad_norm": 0.34633805922704586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130081216748, "tokens_per_second": 152221.25253409377, "data_wait_s": 1.2885000160167692e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2593f8083a02fd6d23e6cac5e8976770768553fbacbef3c3b6e5072763011d7
-size 76413
+oid sha256:3a5d95491e10ef7abffb97fb7743d05a2d842dda214a146042bb422c963c1840
+size 78531