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": 77401.04534051653,
-      "required_pages_per_second_upper_estimate": 0.005354783911825509
+      "required_token_positions_per_second": 77468.36058576666,
+      "required_pages_per_second_upper_estimate": 0.005359440936684801
     },
     {
       "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": 9675.130667564566,
-      "required_pages_per_second_upper_estimate": 0.0040507751708062445
+      "required_token_positions_per_second": 9683.545073220832,
+      "required_pages_per_second_upper_estimate": 0.004054298106741752
     },
     {
       "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": 19350.261335129133,
-      "required_pages_per_second_upper_estimate": 0.004030149135399755
+      "required_token_positions_per_second": 19367.090146441664,
+      "required_pages_per_second_upper_estimate": 0.004033654132990556
     },
     {
       "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": 29025.392002693698,
-      "required_pages_per_second_upper_estimate": 0.0017400279313540645
+      "required_token_positions_per_second": 29050.635219662494,
+      "required_pages_per_second_upper_estimate": 0.0017415412236671832
     },
     {
       "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": 29025.392002693698,
-      "required_pages_per_second_upper_estimate": 0.0022565965590025127
+      "required_token_positions_per_second": 29050.635219662494,
+      "required_pages_per_second_upper_estimate": 0.00225855910808866
     },
     {
       "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": 29025.392002693698,
-      "required_pages_per_second_upper_estimate": 0.37230656357273123
+      "required_token_positions_per_second": 29050.635219662494,
+      "required_pages_per_second_upper_estimate": 0.37263035645595227
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4075755758593118,
-    "producer_p95_seconds": 0.2530779690005147,
-    "gpu_update_seconds": 0.9525452747523317,
-    "producer_batches_per_second": 2.4535326924133036,
-    "gpu_batches_per_second": 1.049818865838169,
-    "producer_headroom": 2.3371009726086585,
+    "producer_batch_seconds": 0.07869178316417447,
+    "producer_p95_seconds": 0.21257635000074515,
+    "gpu_update_seconds": 0.9517175714383986,
+    "producer_batches_per_second": 12.707807089765682,
+    "gpu_batches_per_second": 1.0507318872852465,
+    "producer_headroom": 12.094243301779459,
     "sustainable": true,
-    "buffer_seconds": 60.0103523093969,
+    "buffer_seconds": 60.90992457205751,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374788.4610135123,
-    "upload_bytes_per_second": 5216761.316735143,
+    "generation_bytes_per_second": 374730.32921811915,
+    "upload_bytes_per_second": 5288626.681069306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.42087910465266,
+    "estimated_full_shards_per_hour": 80.40840537460022,
     "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": 1.1199771049996343,
-      "bytes": 6074153,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1042831229997319,
       "bytes": 6145766,
@@ -562,6 +556,12 @@
       "bytes": 6067377,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.3739411900005507,
+      "bytes": 13347857,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -142,3 +142,4 @@
 {"step": 899200, "loss": 0.49028387665748596, "lr": 3e-05, "grad_norm": 0.31439519676396677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131617363438, "tokens_per_second": 154069.68157934587, "data_wait_s": 1.0540999937802553e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899300, "loss": 0.5348533391952515, "lr": 3e-05, "grad_norm": 0.3349686415612178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131631981273, "tokens_per_second": 152231.67781352624, "data_wait_s": 1.5763000192237087e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899400, "loss": 0.48070812225341797, "lr": 3e-05, "grad_norm": 0.3198302746291621, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131646680058, "tokens_per_second": 153482.14705809212, "data_wait_s": 1.5138999515329488e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 899500, "loss": 0.5384294986724854, "lr": 3e-05, "grad_norm": 0.3202262936465282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131661317551, "tokens_per_second": 152490.0874122081, "data_wait_s": 1.6599000446149148e-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 sha256:394ef50e1d437f3b1a20c3f926a04e33119cbeed5da6422e32b4a5393fe9210e
-size 306723
+oid sha256:7029b50e63315a7c231d4e3ab1d3fa99d535296bbe2ed8fa339cf619d34ad072
+size 308841