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": 74324.80415619303,
-      "required_pages_per_second_upper_estimate": 0.0064422856497648815
+      "required_token_positions_per_second": 74718.2088427762,
+      "required_pages_per_second_upper_estimate": 0.006476384970922849
     },
     {
       "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": 9290.600519524129,
-      "required_pages_per_second_upper_estimate": 0.0037061594540945143
+      "required_token_positions_per_second": 9339.776105347026,
+      "required_pages_per_second_upper_estimate": 0.0037257763305197965
     },
     {
       "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": 18581.201039048257,
-      "required_pages_per_second_upper_estimate": 0.004298433842414577
+      "required_token_positions_per_second": 18679.55221069405,
+      "required_pages_per_second_upper_estimate": 0.004321185655053329
     },
     {
       "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": 27871.801558572388,
-      "required_pages_per_second_upper_estimate": 0.0016972464828756881
+      "required_token_positions_per_second": 28019.328316041076,
+      "required_pages_per_second_upper_estimate": 0.0017062300883924537
     },
     {
       "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": 27871.801558572388,
-      "required_pages_per_second_upper_estimate": 0.0021456484380636977
+      "required_token_positions_per_second": 28019.328316041076,
+      "required_pages_per_second_upper_estimate": 0.0021570054562338394
     },
     {
       "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": 27871.801558572388,
-      "required_pages_per_second_upper_estimate": 0.3460744944382382
+      "required_token_positions_per_second": 28019.328316041076,
+      "required_pages_per_second_upper_estimate": 0.3479062830257034
     }
   ],
   "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": 0.12489583282817307,
-    "producer_p95_seconds": 0.33070357799806516,
-    "gpu_update_seconds": 0.9919703231919877,
-    "producer_batches_per_second": 8.006672259239922,
-    "gpu_batches_per_second": 1.0080946744275314,
-    "producer_headroom": 7.942381268690547,
+    "producer_batch_seconds": 0.09313048689068637,
+    "producer_p95_seconds": 0.30637213999943924,
+    "gpu_update_seconds": 0.9867474226415167,
+    "producer_batches_per_second": 10.737622376803081,
+    "gpu_batches_per_second": 1.0134305669864394,
+    "producer_headroom": 10.595321205608318,
     "sustainable": true,
-    "buffer_seconds": 60.51018971471125,
+    "buffer_seconds": 60.19159278113252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401603.2118278261,
-    "upload_bytes_per_second": 5516682.817063272,
+    "generation_bytes_per_second": 385989.2459601814,
+    "upload_bytes_per_second": 5583591.848089599,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.1747004139527,
+    "estimated_full_shards_per_hour": 82.82430681327898,
     "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": 1.0081078959992738,
-      "bytes": 5874669,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0266657669999404,
       "bytes": 4655742,
@@ -562,6 +556,12 @@
       "bytes": 6036267,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.4371028810019197,
+      "bytes": 13333064,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -205,3 +205,4 @@
 {"step": 335500, "loss": 0.5917552970349789, "lr": 0.0003, "grad_norm": 0.14277662475239128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49108894350, "tokens_per_second": 148414.07101879962, "data_wait_s": 1.9260001863585785e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 335600, "loss": 0.5991566870361567, "lr": 0.0003, "grad_norm": 0.16026576510517238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49123424415, "tokens_per_second": 147370.17736312214, "data_wait_s": 1.369399979012087e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 335700, "loss": 0.5827313885092735, "lr": 0.0003, "grad_norm": 0.16445220587395357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49137950562, "tokens_per_second": 147000.08767971842, "data_wait_s": 1.3902001228416339e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 335800, "loss": 0.603604007512331, "lr": 0.0003, "grad_norm": 0.16256666828737928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49152367835, "tokens_per_second": 143218.72598825878, "data_wait_s": 1.428500036126934e-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 sha256:49e66d08d25de319b657fe0bb362fc7baaac994dee662fc48ac310b0932bad42
-size 441723
+oid sha256:d4c7e003f2d9d883c11d7f1969bf10d6d2f7e363f0a28b765cf0723c63243e46
+size 443841