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": 75808.722192994,
-      "required_pages_per_second_upper_estimate": 0.006570907904238909
+      "required_token_positions_per_second": 72989.31603631529,
+      "required_pages_per_second_upper_estimate": 0.006326528924297573
     },
     {
       "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": 9476.09027412425,
-      "required_pages_per_second_upper_estimate": 0.0037801540905234763
+      "required_token_positions_per_second": 9123.66450453941,
+      "required_pages_per_second_upper_estimate": 0.003639566181801265
     },
     {
       "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": 18952.1805482485,
-      "required_pages_per_second_upper_estimate": 0.004384253422851685
+      "required_token_positions_per_second": 18247.32900907882,
+      "required_pages_per_second_upper_estimate": 0.004221198424228188
     },
     {
       "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": 28428.27082237275,
-      "required_pages_per_second_upper_estimate": 0.0017311325414725391
+      "required_token_positions_per_second": 27370.993513618232,
+      "required_pages_per_second_upper_estimate": 0.0016667499004747253
     },
     {
       "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": 28428.27082237275,
-      "required_pages_per_second_upper_estimate": 0.002188486982396562
+      "required_token_positions_per_second": 27370.993513618232,
+      "required_pages_per_second_upper_estimate": 0.0021070948484377297
     },
     {
       "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": 28428.27082237275,
-      "required_pages_per_second_upper_estimate": 0.3529839803118163
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27370.993513618232,
+      "required_pages_per_second_upper_estimate": 0.3398561346166139
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1487863787968422,
-    "producer_p95_seconds": 0.31081133100087754,
-    "gpu_update_seconds": 0.9725529974282261,
-    "producer_batches_per_second": 6.721045354329329,
-    "gpu_batches_per_second": 1.0282216009249403,
-    "producer_headroom": 6.536572805204043,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.298285840550015,
+    "buffer_seconds": 60.607226375419536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380117.94188634434,
-    "upload_bytes_per_second": 4794917.316363702,
+    "generation_bytes_per_second": 292143.268683477,
+    "upload_bytes_per_second": 4907826.024636228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.56446163599728,
+    "estimated_full_shards_per_hour": 62.68714471224053,
     "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": 1.6164238569981535,
-      "bytes": 892499,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.2368712060015241,
       "bytes": 10018327,
@@ -562,6 +556,12 @@
       "bytes": 2646901,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4061265680029464,
+      "bytes": 9691299,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 269200, "loss": 0.5914702340960503, "lr": 0.0003, "grad_norm": 0.14244819826348482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39402884253, "tokens_per_second": 150400.86541815, "data_wait_s": 1.3136999768903479e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:2a591a72f5a3739691ba24de3754ccc593568d402d4e50ef28091fe1f707af76
-size 688977
+oid sha256:40754fcf287b5698fcafa6bfb42eb4521368762cbbd47b32fdde2af65cb675b0
+size 691095