Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75551.94848416465,
-      "required_pages_per_second_upper_estimate": 0.0065486514099445964
+      "required_token_positions_per_second": 75637.51674009152,
+      "required_pages_per_second_upper_estimate": 0.006556068249497576
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9443.993560520581,
-      "required_pages_per_second_upper_estimate": 0.0037673502315783394
+      "required_token_positions_per_second": 9454.68959251144,
+      "required_pages_per_second_upper_estimate": 0.0037716170386594226
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18887.987121041162,
-      "required_pages_per_second_upper_estimate": 0.004369403403233
+      "required_token_positions_per_second": 18909.37918502288,
+      "required_pages_per_second_upper_estimate": 0.004374352080747703
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28331.98068156174,
-      "required_pages_per_second_upper_estimate": 0.001725268977092474
+      "required_token_positions_per_second": 28364.06877753432,
+      "required_pages_per_second_upper_estimate": 0.0017272229737839756
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28331.98068156174,
-      "required_pages_per_second_upper_estimate": 0.002181074300808761
+      "required_token_positions_per_second": 28364.06877753432,
+      "required_pages_per_second_upper_estimate": 0.002183544531262265
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28331.98068156174,
-      "required_pages_per_second_upper_estimate": 0.35178837902531435
+      "required_token_positions_per_second": 28364.06877753432,
+      "required_pages_per_second_upper_estimate": 0.3521868057853449
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07413474426562061,
-    "producer_p95_seconds": 0.19487611200020183,
-    "gpu_update_seconds": 0.9758583528186975,
-    "producer_batches_per_second": 13.48895190650495,
-    "gpu_batches_per_second": 1.0247388846050978,
-    "producer_headroom": 13.163306388732549,
+    "producer_batch_seconds": 0.3330646693436563,
+    "producer_p95_seconds": 0.21111518399993656,
+    "gpu_update_seconds": 0.9747543702862024,
+    "producer_batches_per_second": 3.0024199263482956,
+    "gpu_batches_per_second": 1.025899478354106,
+    "producer_headroom": 2.926621944642379,
     "sustainable": true,
-    "buffer_seconds": 60.503217874759244,
+    "buffer_seconds": 60.43477095774455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338934.65804852644,
-    "upload_bytes_per_second": 3905278.4199953577,
+    "generation_bytes_per_second": 303009.93572211073,
+    "upload_bytes_per_second": 3962121.9681279194,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.72748762218328,
+    "estimated_full_shards_per_hour": 65.01887849566928,
     "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": 0.8003151779998916,
-      "bytes": 1078242,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0970220180001888,
       "bytes": 7456296,
@@ -562,6 +556,12 @@
       "bytes": 2256184,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.5194503539987636,
+      "bytes": 9946744,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -89,3 +89,4 @@
 {"step": 323900, "loss": 0.5672470610588789, "lr": 0.0003, "grad_norm": 0.15293876371581858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47409874860, "tokens_per_second": 89906.31058175236, "data_wait_s": 1.6027999663492665e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324000, "loss": 0.6469105780124664, "lr": 0.0003, "grad_norm": 0.17834168760019067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47424679180, "tokens_per_second": 150837.03804419088, "data_wait_s": 1.5948000509524718e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324100, "loss": 0.5908871721476316, "lr": 0.0003, "grad_norm": 0.17862144998034735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47439485499, "tokens_per_second": 150673.0591248944, "data_wait_s": 2.031700023508165e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 324200, "loss": 0.610282639041543, "lr": 0.0003, "grad_norm": 0.1444129732046015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47453953101, "tokens_per_second": 147405.01491035265, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c56bf9f209bc46d8f4d9999bfa78540395a66f74084b807a43e36228c37b705
-size 193686
+oid sha256:cf16e8798b8430a274f2166329a63ef629788b53103a3ab2887a38aa3ab489df
+size 195804