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": 76700.54940492132,
-      "required_pages_per_second_upper_estimate": 0.0053063219776265285
+      "required_token_positions_per_second": 77231.1318585757,
+      "required_pages_per_second_upper_estimate": 0.005343028902891247
     },
     {
       "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": 9587.568675615164,
-      "required_pages_per_second_upper_estimate": 0.004014114793279348
+      "required_token_positions_per_second": 9653.891482321962,
+      "required_pages_per_second_upper_estimate": 0.004041882767469789
     },
     {
       "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": 19175.13735123033,
-      "required_pages_per_second_upper_estimate": 0.0039936754278836585
+      "required_token_positions_per_second": 19307.782964643924,
+      "required_pages_per_second_upper_estimate": 0.004021302011057648
     },
     {
       "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": 28762.70602684549,
-      "required_pages_per_second_upper_estimate": 0.0017242803082002252
+      "required_token_positions_per_second": 28961.674446965888,
+      "required_pages_per_second_upper_estimate": 0.0017362081611792565
     },
     {
       "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": 28762.70602684549,
-      "required_pages_per_second_upper_estimate": 0.0022361738798138127
+      "required_token_positions_per_second": 28961.674446965888,
+      "required_pages_per_second_upper_estimate": 0.002251642799308584
     },
     {
       "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": 28762.70602684549,
-      "required_pages_per_second_upper_estimate": 0.3689371099247764
+      "required_token_positions_per_second": 28961.674446965888,
+      "required_pages_per_second_upper_estimate": 0.3714892631824359
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.19138600691420038,
-    "producer_p95_seconds": 0.21305518699955428,
-    "gpu_update_seconds": 0.9612447442947445,
-    "producer_batches_per_second": 5.225042395331998,
-    "gpu_batches_per_second": 1.0403177816422704,
-    "producer_headroom": 5.022544541230106,
+    "producer_batch_seconds": 0.08299816250769254,
+    "producer_p95_seconds": 0.23468538199813338,
+    "gpu_update_seconds": 0.9546409359247696,
+    "producer_batches_per_second": 12.048459505441663,
+    "gpu_batches_per_second": 1.0475142667450046,
+    "producer_headroom": 11.501952658726516,
     "sustainable": true,
-    "buffer_seconds": 60.5584188905689,
+    "buffer_seconds": 60.142378963260484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396921.13358982693,
-    "upload_bytes_per_second": 4291767.736136393,
+    "generation_bytes_per_second": 397626.97934179794,
+    "upload_bytes_per_second": 4354110.227863786,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.17003541728121,
+    "estimated_full_shards_per_hour": 85.32149348440603,
     "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": 1.0867800260002696,
-      "bytes": 6181304,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 25.14463806799904,
       "bytes": 7664241,
@@ -562,6 +556,12 @@
       "bytes": 10344850,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3122096120023343,
+      "bytes": 13394029,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 703200, "loss": 0.5644159317016602, "lr": 8e-05, "grad_norm": 0.2075043699022743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102929077700, "tokens_per_second": 153842.47550150743, "data_wait_s": 1.3053999282419682e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703300, "loss": 0.5740440487861633, "lr": 8e-05, "grad_norm": 0.21347681760277257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102943771934, "tokens_per_second": 152922.4717458072, "data_wait_s": 1.7401998775312677e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703400, "loss": 0.5280939936637878, "lr": 8e-05, "grad_norm": 0.20270933958798487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102958507074, "tokens_per_second": 153665.6347216531, "data_wait_s": 1.6848000086611137e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 703500, "loss": 0.49644920229911804, "lr": 8e-05, "grad_norm": 0.19843883597558742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102973085049, "tokens_per_second": 151974.40577166385, "data_wait_s": 1.4781999198021367e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd0a0090efc302d8922dcea79bdd5f6f3c5e977c46d1e5806faaecab476461a5
-size 392226
+oid sha256:38517c5c248eee49c82988e10acf97a097ab0076bd6c7ddf8764eba044acb6cb
+size 394344