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": 75260.9150125208,
-      "required_pages_per_second_upper_estimate": 0.0065234253662400975
+      "required_token_positions_per_second": 75010.43948737466,
+      "required_pages_per_second_upper_estimate": 0.006501714782544845
     },
     {
       "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": 9407.6143765651,
-      "required_pages_per_second_upper_estimate": 0.003752838031181227
+      "required_token_positions_per_second": 9376.304935921833,
+      "required_pages_per_second_upper_estimate": 0.003740348227190774
     },
     {
       "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": 18815.2287531302,
-      "required_pages_per_second_upper_estimate": 0.004352572035320339
+      "required_token_positions_per_second": 18752.609871843666,
+      "required_pages_per_second_upper_estimate": 0.004338086259189371
     },
     {
       "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": 28222.8431296953,
-      "required_pages_per_second_upper_estimate": 0.0017186230727842895
+      "required_token_positions_per_second": 28128.914807765497,
+      "required_pages_per_second_upper_estimate": 0.0017129033307825838
     },
     {
       "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": 28222.8431296953,
-      "required_pages_per_second_upper_estimate": 0.002172672589954003
+      "required_token_positions_per_second": 28128.914807765497,
+      "required_pages_per_second_upper_estimate": 0.00216544172772161
     },
     {
       "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": 28222.8431296953,
-      "required_pages_per_second_upper_estimate": 0.35043325589102275
+      "required_token_positions_per_second": 28128.914807765497,
+      "required_pages_per_second_upper_estimate": 0.34926698049052607
     }
   ],
   "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": 0.8534816499688418,
-    "producer_p95_seconds": 0.3715740870029549,
-    "gpu_update_seconds": 0.9796319907582073,
-    "producer_batches_per_second": 1.171671353492494,
-    "gpu_batches_per_second": 1.0207914905127062,
-    "producer_headroom": 1.1478067405362153,
+    "producer_batch_seconds": 0.1144262393831923,
+    "producer_p95_seconds": 0.3431953010003781,
+    "gpu_update_seconds": 0.9829031865945738,
+    "producer_batches_per_second": 8.73925426012809,
+    "gpu_batches_per_second": 1.0173941987762405,
+    "producer_headroom": 8.589840860740104,
     "sustainable": true,
-    "buffer_seconds": 60.73718342700885,
+    "buffer_seconds": 60.939997568863575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258028.46721506419,
-    "upload_bytes_per_second": 5239844.427358922,
+    "generation_bytes_per_second": 267184.203122032,
+    "upload_bytes_per_second": 5299534.799902086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.36690246905274,
+    "estimated_full_shards_per_hour": 57.33151026006431,
     "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": 1.1865619210002478,
-      "bytes": 5319050,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3752340050014027,
       "bytes": 13115816,
@@ -562,6 +556,12 @@
       "bytes": 7394502,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2961700559972087,
+      "bytes": 10630748,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -360,3 +360,4 @@
 {"step": 273100, "loss": 0.5965057983994484, "lr": 0.0003, "grad_norm": 0.16545406222730363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39973025471, "tokens_per_second": 150156.61626166187, "data_wait_s": 1.3949000276625156e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273200, "loss": 0.6044473890215158, "lr": 0.0003, "grad_norm": 0.1546214700058153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39987728968, "tokens_per_second": 150533.62304358085, "data_wait_s": 1.3463999493978918e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273300, "loss": 0.6495624426752329, "lr": 0.0003, "grad_norm": 0.16367195483570454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40002486610, "tokens_per_second": 118874.44948911833, "data_wait_s": 1.1262993211857975e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 273400, "loss": 0.5447521097958088, "lr": 0.0003, "grad_norm": 0.1448215740828469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40017066899, "tokens_per_second": 148404.4334698735, "data_wait_s": 9.012997907120734e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ddee7478d3639b059c512ae47928733eb1dfedfe8d1df0fbbafb69c51ff1b63f
-size 772362
+oid sha256:295905491b2970129f5547db53c3694823a907fa1823d7dd4436aeed25259802
+size 774480