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": 75196.39230722244,
-      "required_pages_per_second_upper_estimate": 0.006517832701676138
+      "required_token_positions_per_second": 74077.8125041623,
+      "required_pages_per_second_upper_estimate": 0.006420877039361476
     },
     {
       "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": 9399.549038402805,
-      "required_pages_per_second_upper_estimate": 0.003749620647200736
+      "required_token_positions_per_second": 9259.726563020287,
+      "required_pages_per_second_upper_estimate": 0.0036938433712383465
     },
     {
       "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": 18799.09807680561,
-      "required_pages_per_second_upper_estimate": 0.004348840487242853
+      "required_token_positions_per_second": 18519.453126040575,
+      "required_pages_per_second_upper_estimate": 0.004284149549466403
     },
     {
       "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": 28198.64711520842,
-      "required_pages_per_second_upper_estimate": 0.0017171496624487157
+      "required_token_positions_per_second": 27779.17968906086,
+      "required_pages_per_second_upper_estimate": 0.0016916062969718293
     },
     {
       "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": 28198.64711520842,
-      "required_pages_per_second_upper_estimate": 0.0021708099137799476
+      "required_token_positions_per_second": 27779.17968906086,
+      "required_pages_per_second_upper_estimate": 0.0021385181501549556
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28198.64711520842,
-      "required_pages_per_second_upper_estimate": 0.35013282236994697
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27779.17968906086,
+      "required_pages_per_second_upper_estimate": 0.34492444080436147
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "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,
+    "producer_batch_seconds": 0.5326533081015157,
+    "producer_p95_seconds": 0.27997057199900155,
+    "gpu_update_seconds": 0.9952777695191439,
+    "producer_batches_per_second": 1.877393765870342,
+    "gpu_batches_per_second": 1.0047446357443888,
+    "producer_headroom": 1.8685282798045797,
     "sustainable": true,
-    "buffer_seconds": 60.78929932335267,
+    "buffer_seconds": 60.71194394066778,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308191.60098377947,
-    "upload_bytes_per_second": 5207608.577218202,
+    "generation_bytes_per_second": 267724.8063110895,
+    "upload_bytes_per_second": 5232798.352871692,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.13074323782958,
+    "estimated_full_shards_per_hour": 57.44751111983789,
     "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": 2.8337960279986874,
-      "bytes": 11685666,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.282873722999284,
       "bytes": 2666361,
@@ -562,6 +556,12 @@
       "bytes": 1139782,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.287253474998579,
+      "bytes": 5620920,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -189,3 +189,4 @@
 {"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}
+{"step": 256300, "loss": 0.5564228966832161, "lr": 0.0003, "grad_norm": 0.14486568315783913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37515491507, "tokens_per_second": 124273.96550886951, "data_wait_s": 2.0496001525316387e-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 sha256:181cc1f8e83ed1abd99d6eef75d994000809f492ea034ea6ded06b7dddc2e116
-size 407835
+oid sha256:cb8ee68cba0369a9ccd14fd3ca48002d59316242d932eda2c71482d653db3df0
+size 409953