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": 75288.04890585551,
-      "required_pages_per_second_upper_estimate": 0.006525777263344127
+      "required_token_positions_per_second": 75196.39230722244,
+      "required_pages_per_second_upper_estimate": 0.006517832701676138
     },
     {
       "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": 9411.006113231939,
-      "required_pages_per_second_upper_estimate": 0.0037541910456486112
+      "required_token_positions_per_second": 9399.549038402805,
+      "required_pages_per_second_upper_estimate": 0.003749620647200736
     },
     {
       "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": 18822.012226463878,
-      "required_pages_per_second_upper_estimate": 0.004354141272491033
+      "required_token_positions_per_second": 18799.09807680561,
+      "required_pages_per_second_upper_estimate": 0.004348840487242853
     },
     {
       "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": 28233.01833969582,
-      "required_pages_per_second_upper_estimate": 0.0017192426896881201
+      "required_token_positions_per_second": 28198.64711520842,
+      "required_pages_per_second_upper_estimate": 0.0017171496624487157
     },
     {
       "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": 28233.01833969582,
-      "required_pages_per_second_upper_estimate": 0.002173455905786627
+      "required_token_positions_per_second": 28198.64711520842,
+      "required_pages_per_second_upper_estimate": 0.0021708099137799476
     },
     {
       "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": 28233.01833969582,
-      "required_pages_per_second_upper_estimate": 0.350559597944992
+      "required_token_positions_per_second": 28198.64711520842,
+      "required_pages_per_second_upper_estimate": 0.35013282236994697
     }
   ],
   "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": 0.08908820581277155,
-    "producer_p95_seconds": 0.22685217899925192,
-    "gpu_update_seconds": 0.9792789303411715,
-    "producer_batches_per_second": 11.224830390025001,
-    "gpu_batches_per_second": 1.0211595174947852,
-    "producer_headroom": 10.992239897604758,
+    "producer_batch_seconds": 0.09065544070318765,
+    "producer_p95_seconds": 0.21573033300228417,
+    "gpu_update_seconds": 0.9804725697314948,
+    "producer_batches_per_second": 11.03077754896224,
+    "gpu_batches_per_second": 1.0199163453128044,
+    "producer_headroom": 10.815374809567487,
     "sustainable": true,
-    "buffer_seconds": 60.71529368115264,
+    "buffer_seconds": 60.78929932335267,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296380.5872115805,
-    "upload_bytes_per_second": 5247246.010994968,
+    "generation_bytes_per_second": 308191.60098377947,
+    "upload_bytes_per_second": 5207608.577218202,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.59637462864457,
+    "estimated_full_shards_per_hour": 66.13074323782958,
     "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": 1.2046007990011276,
-      "bytes": 5154137,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 2.8337960279986874,
       "bytes": 11685666,
@@ -562,6 +556,12 @@
       "bytes": 13143755,
       "pages": 31,
       "shards": 4
+    },
+    {
+      "seconds": 1.047720095000841,
+      "bytes": 1139782,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -188,3 +188,4 @@
 {"step": 255900, "loss": 0.6159732881933451, "lr": 0.0003, "grad_norm": 0.1648850653482918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37457210876, "tokens_per_second": 151658.94846547183, "data_wait_s": 1.2847998732468113e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256000, "loss": 0.5724415425211191, "lr": 0.0003, "grad_norm": 0.15820589177725958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37471774801, "tokens_per_second": 147479.50363534992, "data_wait_s": 1.5329998859670013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256100, "loss": 0.6006165221333504, "lr": 0.0003, "grad_norm": 0.16127717517682835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37486250693, "tokens_per_second": 147841.50769142716, "data_wait_s": 1.4294000720838085e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 256200, "loss": 0.6199516355991364, "lr": 0.0003, "grad_norm": 0.14335172639052332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37500863935, "tokens_per_second": 149267.98370844394, "data_wait_s": 1.6933001461438835e-05, "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 sha256:216a380adaf7e0f6a45e819bd99f159b129086b361f498d1720b4d9b14b380fb
-size 405717
+oid sha256:181cc1f8e83ed1abd99d6eef75d994000809f492ea034ea6ded06b7dddc2e116
+size 407835