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": 76743.47490185667,
-      "required_pages_per_second_upper_estimate": 0.005309291663105392
+      "required_token_positions_per_second": 76844.82269293365,
+      "required_pages_per_second_upper_estimate": 0.0053163031384513715
     },
     {
       "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": 9592.934362732083,
-      "required_pages_per_second_upper_estimate": 0.004016361294426913
+      "required_token_positions_per_second": 9605.602836616707,
+      "required_pages_per_second_upper_estimate": 0.004021665319894588
     },
     {
       "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": 19185.868725464166,
-      "required_pages_per_second_upper_estimate": 0.003995910490131197
+      "required_token_positions_per_second": 19211.205673233413,
+      "required_pages_per_second_upper_estimate": 0.004001187508171285
     },
     {
       "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": 28778.80308819625,
-      "required_pages_per_second_upper_estimate": 0.0017252453024494139
+      "required_token_positions_per_second": 28816.808509850118,
+      "required_pages_per_second_upper_estimate": 0.0017275236694466443
     },
     {
       "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": 28778.80308819625,
-      "required_pages_per_second_upper_estimate": 0.0022374253555303442
+      "required_token_positions_per_second": 28816.808509850118,
+      "required_pages_per_second_upper_estimate": 0.0022403801098957493
     },
     {
       "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": 28778.80308819625,
-      "required_pages_per_second_upper_estimate": 0.36914358574410605
+      "required_token_positions_per_second": 28816.808509850118,
+      "required_pages_per_second_upper_estimate": 0.3696310784860394
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.26605430171090916,
-    "producer_p95_seconds": 0.2448797530014417,
-    "gpu_update_seconds": 0.9607070841434662,
-    "producer_batches_per_second": 3.758631202612863,
-    "gpu_batches_per_second": 1.0408999959561722,
-    "producer_headroom": 3.610943623032853,
+    "producer_batch_seconds": 0.0661616753671126,
+    "producer_p95_seconds": 0.24438537600144628,
+    "gpu_update_seconds": 0.9594400431452845,
+    "producer_batches_per_second": 15.11449029141539,
+    "gpu_batches_per_second": 1.0422746133481668,
+    "producer_headroom": 14.501447217314565,
     "sustainable": true,
-    "buffer_seconds": 60.52454630103837,
+    "buffer_seconds": 60.444722718152924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406749.66645443335,
-    "upload_bytes_per_second": 5390841.985321921,
+    "generation_bytes_per_second": 406720.83803339687,
+    "upload_bytes_per_second": 5366265.038987921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.27900977349043,
+    "estimated_full_shards_per_hour": 87.2728238654273,
     "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": 1.2397414250008296,
-      "bytes": 13176621,
-      "pages": 29,
-      "shards": 4
-    },
     {
       "seconds": 1.2003755249988899,
       "bytes": 10273667,
@@ -562,6 +556,12 @@
       "bytes": 2394266,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1657910530011577,
+      "bytes": 10819661,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 579700, "loss": 0.5969656109809875, "lr": 0.0003, "grad_norm": 0.13797264914020052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84850919354, "tokens_per_second": 152031.74969802043, "data_wait_s": 9.548999514663592e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 579800, "loss": 0.6223017573356628, "lr": 0.0003, "grad_norm": 0.15240272493459464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84865539341, "tokens_per_second": 152289.9581169443, "data_wait_s": 1.3045999367022887e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579900, "loss": 0.5686430931091309, "lr": 0.0003, "grad_norm": 0.16859840235040152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84880212579, "tokens_per_second": 153153.92990083736, "data_wait_s": 8.338996849488467e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 580000, "loss": 0.6204338073730469, "lr": 0.0003, "grad_norm": 0.17000273589656398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84894863938, "tokens_per_second": 152097.15350274555, "data_wait_s": 1.1026997526641935e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01f058d46c04d27bd91a3b1302d6cd72e543c81433508356b8d7dd27cd8f2260
-size 486201
+oid sha256:9da48cba1521df3e3a1be4656063f1c4b63468e1b54db67917e10df2df03ca42
+size 488319