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": 74535.5237026849,
-      "required_pages_per_second_upper_estimate": 0.006460550285990989
+      "required_token_positions_per_second": 74915.79870669916,
+      "required_pages_per_second_upper_estimate": 0.006493511559540759
     },
     {
       "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": 9316.940462835613,
-      "required_pages_per_second_upper_estimate": 0.003716666851298713
+      "required_token_positions_per_second": 9364.474838337395,
+      "required_pages_per_second_upper_estimate": 0.003735629024388621
     },
     {
       "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": 18633.880925671227,
-      "required_pages_per_second_upper_estimate": 0.004310620407050462
+      "required_token_positions_per_second": 18728.94967667479,
+      "required_pages_per_second_upper_estimate": 0.004332612889442265
     },
     {
       "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": 27950.82138850684,
-      "required_pages_per_second_upper_estimate": 0.0017020583759336896
+      "required_token_positions_per_second": 28093.42451501219,
+      "required_pages_per_second_upper_estimate": 0.001710742158157083
     },
     {
       "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": 27950.82138850684,
-      "required_pages_per_second_upper_estimate": 0.0021517316033129413
+      "required_token_positions_per_second": 28093.42451501219,
+      "required_pages_per_second_upper_estimate": 0.002162709586742042
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27950.82138850684,
-      "required_pages_per_second_upper_estimate": 0.34705565626366564
+      "required_token_positions_per_second": 28093.42451501219,
+      "required_pages_per_second_upper_estimate": 0.34882630983289903
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5027312837500801,
-    "producer_p95_seconds": 0.30342336400281056,
-    "gpu_update_seconds": 0.9891659216629907,
-    "producer_batches_per_second": 1.989134220056065,
-    "gpu_batches_per_second": 1.0109527411931003,
-    "producer_headroom": 1.9675837840931516,
+    "producer_batch_seconds": 0.11260153542960438,
+    "producer_p95_seconds": 0.2763736630004132,
+    "gpu_update_seconds": 0.9841448836266234,
+    "producer_batches_per_second": 8.880873570548907,
+    "gpu_batches_per_second": 1.0161105510348738,
+    "producer_headroom": 8.740066286590611,
     "sustainable": true,
-    "buffer_seconds": 60.33912122144243,
+    "buffer_seconds": 60.03283790122403,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283647.0919123367,
-    "upload_bytes_per_second": 5556239.060564422,
+    "generation_bytes_per_second": 284294.68287694437,
+    "upload_bytes_per_second": 5510991.289085294,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.86406295802666,
+    "estimated_full_shards_per_hour": 61.00302090388535,
     "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": 1.2733968520005874,
-      "bytes": 11422716,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.048131932002434,
       "bytes": 6792716,
@@ -562,6 +556,12 @@
       "bytes": 5754870,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1586719139995694,
+      "bytes": 7299186,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -236,3 +236,4 @@
 {"step": 338600, "loss": 0.6005110405385494, "lr": 0.0003, "grad_norm": 0.19402572089345949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49561984563, "tokens_per_second": 147040.83739049642, "data_wait_s": 1.6848000086611137e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338700, "loss": 0.6105085415765643, "lr": 0.0003, "grad_norm": 0.16156311645911683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49576486868, "tokens_per_second": 135554.30426191163, "data_wait_s": 1.5159999747993425e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338800, "loss": 0.6380332540720701, "lr": 0.0003, "grad_norm": 0.16096344868583667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49591185901, "tokens_per_second": 148972.36315423885, "data_wait_s": 1.1410000297473744e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 338900, "loss": 0.5484571140259504, "lr": 0.0003, "grad_norm": 0.1442884114183441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49605719347, "tokens_per_second": 147974.9637834689, "data_wait_s": 1.793300179997459e-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 sha256:17647e5629a0f9891a804cbd23ffc124625f73c690e8879dbe593b8e6414cfb1
-size 507381
+oid sha256:47ef5b48c4c02fb0e624b6137d9dc99b453a1141258722074519b438b49ae58d
+size 509499