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": 75898.99250477127,
-      "required_pages_per_second_upper_estimate": 0.00657873230607575
+      "required_token_positions_per_second": 76108.63377546435,
+      "required_pages_per_second_upper_estimate": 0.006596903480088482
     },
     {
       "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": 9487.37406309641,
-      "required_pages_per_second_upper_estimate": 0.00378465536265215
+      "required_token_positions_per_second": 9513.579221933043,
+      "required_pages_per_second_upper_estimate": 0.003795108992314123
     },
     {
       "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": 18974.74812619282,
-      "required_pages_per_second_upper_estimate": 0.004389474034833297
+      "required_token_positions_per_second": 19027.158443866087,
+      "required_pages_per_second_upper_estimate": 0.0044015982394369196
     },
     {
       "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": 28462.122189289224,
-      "required_pages_per_second_upper_estimate": 0.0017331939121133563
+      "required_token_positions_per_second": 28540.73766579913,
+      "required_pages_per_second_upper_estimate": 0.0017379811821692812
     },
     {
       "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": 28462.122189289224,
-      "required_pages_per_second_upper_estimate": 0.0021910929543283732
+      "required_token_positions_per_second": 28540.73766579913,
+      "required_pages_per_second_upper_estimate": 0.0021971449913316733
     },
     {
       "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": 28462.122189289224,
-      "required_pages_per_second_upper_estimate": 0.3534043009956818
+      "required_token_positions_per_second": 28540.73766579913,
+      "required_pages_per_second_upper_estimate": 0.35438044210486025
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08998943196880305,
-    "producer_p95_seconds": 0.35661485899981926,
-    "gpu_update_seconds": 0.9713962935063362,
-    "producer_batches_per_second": 11.112415959539266,
-    "gpu_batches_per_second": 1.0294459703880652,
-    "producer_headroom": 10.7945596749971,
+    "producer_batch_seconds": 0.10494775075002849,
+    "producer_p95_seconds": 0.351025005998963,
+    "gpu_update_seconds": 0.9687205819186336,
+    "producer_batches_per_second": 9.528551044241684,
+    "gpu_batches_per_second": 1.0322894121021098,
+    "producer_headroom": 9.230503512419208,
     "sustainable": true,
-    "buffer_seconds": 60.22657019739284,
+    "buffer_seconds": 60.06067607895528,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 676949.0283745778,
-    "upload_bytes_per_second": 5120259.387260101,
+    "generation_bytes_per_second": 685546.3634471458,
+    "upload_bytes_per_second": 5172623.613788709,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.62875146116257,
+    "estimated_full_shards_per_hour": 73.55114544659033,
     "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": 1.1722596879990306,
-      "bytes": 1208158,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2069239850025042,
       "bytes": 7251816,
@@ -562,6 +556,12 @@
       "bytes": 10157591,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.351508649000607,
+      "bytes": 6405900,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -301,3 +301,4 @@
 {"step": 145100, "loss": 0.5617381744086742, "lr": 0.0003, "grad_norm": 0.12929500746264588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21235033207, "tokens_per_second": 139643.53645174686, "data_wait_s": 2.4274999304907396e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145200, "loss": 0.6110075414180756, "lr": 0.0003, "grad_norm": 0.13060606311302397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21249704512, "tokens_per_second": 149717.61129311027, "data_wait_s": 1.4770001143915579e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145300, "loss": 0.610079487785697, "lr": 0.0003, "grad_norm": 0.1429218621358837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21264470418, "tokens_per_second": 151096.84181564505, "data_wait_s": 1.37290007842239e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 145400, "loss": 0.6292114295065403, "lr": 0.0003, "grad_norm": 0.1463837511759007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21279195490, "tokens_per_second": 151062.97558532076, "data_wait_s": 1.0263000149279833e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a00bc36ee85e1d8ad849b95d6d6c009f25a88ece476ce383b09ad6c9f016e239
-size 646617
+oid sha256:16c7c1a128852bf6dd918a8fc17fc24e04cf3f8e4b86b6a0c0fe8175e5865fd3
+size 648735