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": 75714.50405596355,
-      "required_pages_per_second_upper_estimate": 0.005238105072445204
+      "required_token_positions_per_second": 77435.7584107616,
+      "required_pages_per_second_upper_estimate": 0.005357185442570482
     },
     {
       "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": 9464.313006995444,
-      "required_pages_per_second_upper_estimate": 0.003962510218699316
+      "required_token_positions_per_second": 9679.4698013452,
+      "required_pages_per_second_upper_estimate": 0.004052591875508779
     },
     {
       "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": 18928.626013990888,
-      "required_pages_per_second_upper_estimate": 0.003942333617277816
+      "required_token_positions_per_second": 19358.9396026904,
+      "required_pages_per_second_upper_estimate": 0.004031956589671461
     },
     {
       "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": 28392.93902098633,
-      "required_pages_per_second_upper_estimate": 0.001702113340800495
+      "required_token_positions_per_second": 29038.409404035596,
+      "required_pages_per_second_upper_estimate": 0.0017408083046880902
     },
     {
       "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": 28392.93902098633,
-      "required_pages_per_second_upper_estimate": 0.0022074261215414346
+      "required_token_positions_per_second": 29038.409404035596,
+      "required_pages_per_second_upper_estimate": 0.0022576086046993482
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28392.93902098633,
-      "required_pages_per_second_upper_estimate": 0.36419413579849325
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29038.409404035596,
+      "required_pages_per_second_upper_estimate": 0.3724735368201485
     }
   ],
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.312966403265591,
-    "producer_p95_seconds": 0.29335017899938975,
-    "gpu_update_seconds": 0.9737632296383366,
-    "producer_batches_per_second": 3.1952311480263758,
-    "gpu_batches_per_second": 1.0269436856548875,
-    "producer_headroom": 3.111398602143174,
+    "producer_batch_seconds": 0.1313207685232669,
+    "producer_p95_seconds": 0.3081663520024449,
+    "gpu_update_seconds": 0.9521182656842642,
+    "producer_batches_per_second": 7.614941728145795,
+    "gpu_batches_per_second": 1.0502896919862412,
+    "producer_headroom": 7.250325111488908,
     "sustainable": true,
-    "buffer_seconds": 60.37332023757687,
+    "buffer_seconds": 60.93556900379291,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408927.86943115393,
-    "upload_bytes_per_second": 5373547.223077975,
+    "generation_bytes_per_second": 425021.911897896,
+    "upload_bytes_per_second": 5366755.264876879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.7464014263245,
+    "estimated_full_shards_per_hour": 91.19980828955326,
     "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": 1.1900975380012824,
-      "bytes": 11578816,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.276093910997588,
       "bytes": 903821,
@@ -562,6 +556,12 @@
       "bytes": 5705052,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.4537645320015145,
+      "bytes": 12446484,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -286,3 +286,4 @@
 {"step": 673600, "loss": 0.5435763001441956, "lr": 0.0001, "grad_norm": 0.2397423740731034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98597483724, "tokens_per_second": 150260.20550584595, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673700, "loss": 0.5673876404762268, "lr": 0.0001, "grad_norm": 0.2134560078421933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98611932000, "tokens_per_second": 150182.8946675715, "data_wait_s": 1.2245000107213855e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673800, "loss": 0.5613048076629639, "lr": 0.0001, "grad_norm": 0.21319297970065137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98626528191, "tokens_per_second": 152388.1773700583, "data_wait_s": 1.0379997547715902e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 673900, "loss": 0.5678796768188477, "lr": 0.0001, "grad_norm": 0.19500271852505002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98641213613, "tokens_per_second": 152612.64458449662, "data_wait_s": 1.095100014936179e-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:c6a46a953ff8ca44dd90c6a6a21818c7c681d7e86a3541ad4b5e0be27ef779ca
-size 614064
+oid sha256:0f113145000f12dae56821ab849412118bc3c59d779e11063ab25b082d883772
+size 616182