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": 76701.5083632127,
-      "required_pages_per_second_upper_estimate": 0.005306388320586217
+      "required_token_positions_per_second": 76152.1819837258,
+      "required_pages_per_second_upper_estimate": 0.005268384647040505
     },
     {
       "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": 9587.688545401588,
-      "required_pages_per_second_upper_estimate": 0.004014164980255757
+      "required_token_positions_per_second": 9519.022747965724,
+      "required_pages_per_second_upper_estimate": 0.003985416044774267
     },
     {
       "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": 19175.377090803177,
-      "required_pages_per_second_upper_estimate": 0.003993725359314325
+      "required_token_positions_per_second": 19038.04549593145,
+      "required_pages_per_second_upper_estimate": 0.003965122809780248
     },
     {
       "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": 28763.065636204767,
-      "required_pages_per_second_upper_estimate": 0.0017243018662322222
+      "required_token_positions_per_second": 28557.06824389717,
+      "required_pages_per_second_upper_estimate": 0.0017119526371031874
     },
     {
       "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": 28763.065636204767,
-      "required_pages_per_second_upper_estimate": 0.002236201837859742
+      "required_token_positions_per_second": 28557.06824389717,
+      "required_pages_per_second_upper_estimate": 0.002220186446694597
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28763.065636204767,
-      "required_pages_per_second_upper_estimate": 0.36894172260751873
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28557.06824389717,
+      "required_pages_per_second_upper_estimate": 0.3662994092417641
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10364180496092246,
-    "producer_p95_seconds": 0.15860815399992134,
-    "gpu_update_seconds": 0.9612327263613651,
-    "producer_batches_per_second": 9.648616215985859,
-    "gpu_batches_per_second": 1.0403307883465265,
-    "producer_headroom": 9.274565670906565,
+    "producer_batch_seconds": 0.05107470774220957,
+    "producer_p95_seconds": 0.16934688499986805,
+    "gpu_update_seconds": 0.9681666116376829,
+    "producer_batches_per_second": 19.579162450567917,
+    "gpu_batches_per_second": 1.0328800724789198,
+    "producer_headroom": 18.95589136847009,
     "sustainable": true,
-    "buffer_seconds": 60.557661760766,
+    "buffer_seconds": 60.02632992153634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442669.4485731921,
-    "upload_bytes_per_second": 1327168.3025109041,
+    "generation_bytes_per_second": 441519.67414723465,
+    "upload_bytes_per_second": 1400953.169869347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.98655884644339,
+    "estimated_full_shards_per_hour": 94.7398440200117,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 2497791,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.873195456000758,
+      "bytes": 6027495,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -44,3 +44,4 @@
 {"step": 649400, "loss": 0.5353020429611206, "lr": 0.0001, "grad_norm": 0.19023563998215592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95055694777, "tokens_per_second": 153860.38491552442, "data_wait_s": 1.1023000297427643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 649500, "loss": 0.5365878939628601, "lr": 0.0001, "grad_norm": 0.19241890259132047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95070355589, "tokens_per_second": 153102.34730799394, "data_wait_s": 8.099000297079328e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 649600, "loss": 0.5479921698570251, "lr": 0.0001, "grad_norm": 0.19101405004114777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95084832627, "tokens_per_second": 150693.50332473035, "data_wait_s": 9.286000022257213e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 649700, "loss": 0.5493369102478027, "lr": 0.0001, "grad_norm": 0.16725687475794224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95099487377, "tokens_per_second": 152387.97609152904, "data_wait_s": 1.0320999535906594e-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 sha256:31fb175794a8f573a31d85a8a7d2a47aab43b1bdab46e6c6d755c1e09176c438
-size 97593
+oid sha256:52800e9f3f5d9256212f2f66762ce2596f836fd5b0b97f77254b8c78b08a8968
+size 99711