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": 72989.31603631529,
-      "required_pages_per_second_upper_estimate": 0.006326528924297573
+      "required_token_positions_per_second": 75718.21135527581,
+      "required_pages_per_second_upper_estimate": 0.006563062654223117
     },
     {
       "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": 9123.66450453941,
-      "required_pages_per_second_upper_estimate": 0.003639566181801265
+      "required_token_positions_per_second": 9464.776419409476,
+      "required_pages_per_second_upper_estimate": 0.0037756408247205508
     },
     {
       "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": 18247.32900907882,
-      "required_pages_per_second_upper_estimate": 0.004221198424228188
+      "required_token_positions_per_second": 18929.552838818952,
+      "required_pages_per_second_upper_estimate": 0.004379018900509244
     },
     {
       "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": 27370.993513618232,
-      "required_pages_per_second_upper_estimate": 0.0016667499004747253
+      "required_token_positions_per_second": 28394.329258228427,
+      "required_pages_per_second_upper_estimate": 0.0017290656782937746
     },
     {
       "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": 27370.993513618232,
-      "required_pages_per_second_upper_estimate": 0.0021070948484377297
+      "required_token_positions_per_second": 28394.329258228427,
+      "required_pages_per_second_upper_estimate": 0.0021858740668324692
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27370.993513618232,
-      "required_pages_per_second_upper_estimate": 0.3398561346166139
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28394.329258228427,
+      "required_pages_per_second_upper_estimate": 0.35256253968025164
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.6877988105862869,
-    "producer_p95_seconds": 0.3474228580016643,
-    "gpu_update_seconds": 1.0101204395903256,
-    "producer_batches_per_second": 1.4539135349006922,
-    "gpu_batches_per_second": 0.9899809575238077,
-    "producer_headroom": 1.4686277790002114,
+    "producer_batch_seconds": 0.12386094254702584,
+    "producer_p95_seconds": 0.33616301199799636,
+    "gpu_update_seconds": 0.9737155524456649,
+    "producer_batches_per_second": 8.0735700813865,
+    "gpu_batches_per_second": 1.0269939691199519,
+    "producer_headroom": 7.861360752006047,
     "sustainable": true,
-    "buffer_seconds": 60.607226375419536,
+    "buffer_seconds": 60.37036425163122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292143.268683477,
-    "upload_bytes_per_second": 4907826.024636228,
+    "generation_bytes_per_second": 291062.10054246854,
+    "upload_bytes_per_second": 4882413.032388977,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.68714471224053,
+    "estimated_full_shards_per_hour": 62.45515119748632,
     "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": 1.2368712060015241,
-      "bytes": 10018327,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.856085086998064,
       "bytes": 7579977,
@@ -562,6 +556,12 @@
       "bytes": 9691299,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.0368577580011333,
+      "bytes": 6834427,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -322,3 +322,4 @@
 {"step": 269300, "loss": 0.612301142886281, "lr": 0.0003, "grad_norm": 0.15935870466315966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39417600715, "tokens_per_second": 150237.53706805658, "data_wait_s": 2.0755000150529668e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269400, "loss": 0.6281353235244751, "lr": 0.0003, "grad_norm": 0.19009981058232467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39432218415, "tokens_per_second": 149537.97952637076, "data_wait_s": 1.4057997759664431e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269500, "loss": 0.6397731490433216, "lr": 0.0003, "grad_norm": 0.13954207191130794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39446894944, "tokens_per_second": 107512.69546627633, "data_wait_s": 1.5251003787852824e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 269600, "loss": 0.6128731053322554, "lr": 0.0003, "grad_norm": 0.1641111567820474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39461615948, "tokens_per_second": 150870.7711908302, "data_wait_s": 1.6453996067866683e-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 sha256:40754fcf287b5698fcafa6bfb42eb4521368762cbbd47b32fdde2af65cb675b0
-size 691095
+oid sha256:a98b974cee6db6cb1b447208e5a0a7401ff1fdbe798d5d2a8c4137f184668bea
+size 693213