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": 77577.90939393648,
-      "required_pages_per_second_upper_estimate": 0.005367019777422244
+      "required_token_positions_per_second": 76732.07735116074,
+      "required_pages_per_second_upper_estimate": 0.0053085031541022865
     },
     {
       "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": 9697.23867424206,
-      "required_pages_per_second_upper_estimate": 0.004060031331534434
+      "required_token_positions_per_second": 9591.509668895093,
+      "required_pages_per_second_upper_estimate": 0.004015764804868356
     },
     {
       "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": 19394.47734848412,
-      "required_pages_per_second_upper_estimate": 0.004039358164926913
+      "required_token_positions_per_second": 19183.019337790185,
+      "required_pages_per_second_upper_estimate": 0.003995317037822113
     },
     {
       "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": 29091.71602272618,
-      "required_pages_per_second_upper_estimate": 0.001744003955084097
+      "required_token_positions_per_second": 28774.529006685276,
+      "required_pages_per_second_upper_estimate": 0.0017249890777889755
     },
     {
       "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": 29091.71602272618,
-      "required_pages_per_second_upper_estimate": 0.002261752959831505
+      "required_token_positions_per_second": 28774.529006685276,
+      "required_pages_per_second_upper_estimate": 0.0022370930644925635
     },
     {
       "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": 29091.71602272618,
-      "required_pages_per_second_upper_estimate": 0.3731572968885235
+      "required_token_positions_per_second": 28774.529006685276,
+      "required_pages_per_second_upper_estimate": 0.36908876241563443
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1407548160155443,
-    "producer_p95_seconds": 0.332064309004636,
-    "gpu_update_seconds": 0.9503736382687648,
-    "producer_batches_per_second": 7.104552642018052,
-    "gpu_batches_per_second": 1.0522177380905013,
-    "producer_headroom": 6.751979542666661,
+    "producer_batch_seconds": 0.10379790004657252,
+    "producer_p95_seconds": 0.30956860299920663,
+    "gpu_update_seconds": 0.9608497846681158,
+    "producer_batches_per_second": 9.634106273357316,
+    "gpu_batches_per_second": 1.0407454067811515,
+    "producer_headroom": 9.256928938225121,
     "sustainable": true,
-    "buffer_seconds": 60.823912849200944,
+    "buffer_seconds": 60.533536434091296,
     "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": 457494.1428036806,
-    "upload_bytes_per_second": 5046843.635732153,
+    "generation_bytes_per_second": 493037.7153887816,
+    "upload_bytes_per_second": 5056775.41149406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.16759312708677,
+    "estimated_full_shards_per_hour": 105.7944163918265,
     "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": 1.4761009860012564,
-      "bytes": 5975966,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.119130936000147,
       "bytes": 7848522,
@@ -562,6 +556,12 @@
       "bytes": 7352671,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.3506424060033169,
+      "bytes": 6203770,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 683100, "loss": 0.540785551071167, "lr": 0.0001, "grad_norm": 0.2015798868947121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99988027993, "tokens_per_second": 151482.1991421006, "data_wait_s": 8.834998880047351e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 683200, "loss": 0.5725772380828857, "lr": 0.0001, "grad_norm": 0.2014802505878475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100002652287, "tokens_per_second": 153065.17170915345, "data_wait_s": 1.7001002561300993e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683300, "loss": 0.5685223937034607, "lr": 0.0001, "grad_norm": 0.20739122870950957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100017359825, "tokens_per_second": 152272.69776061346, "data_wait_s": 1.486000110162422e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683400, "loss": 0.5565347075462341, "lr": 0.0001, "grad_norm": 0.1834445118323151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100032050677, "tokens_per_second": 152632.3169792421, "data_wait_s": 1.0711002687457949e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0325e10f4bd210cbc736a4dcadcd65710c3a2e06f82d68da27fb1294dc667093
-size 816840
+oid sha256:7fdbfcafde01ecbbf4a4522a78a66f477a131dd9d14a25e35d52871a69303359
+size 818958