Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76846.3169455834,
-      "required_pages_per_second_upper_estimate": 0.0053164065143168245
+      "required_token_positions_per_second": 75590.43250265361,
+      "required_pages_per_second_upper_estimate": 0.005229521514501556
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9605.789618197925,
-      "required_pages_per_second_upper_estimate": 0.0040217435214422005
+      "required_token_positions_per_second": 9448.804062831701,
+      "required_pages_per_second_upper_estimate": 0.00395601694764154
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19211.57923639585,
-      "required_pages_per_second_upper_estimate": 0.004001265311526498
+      "required_token_positions_per_second": 18897.608125663402,
+      "required_pages_per_second_upper_estimate": 0.003935873409135923
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28817.368854593773,
-      "required_pages_per_second_upper_estimate": 0.0017275572612584296
+      "required_token_positions_per_second": 28346.412188495106,
+      "required_pages_per_second_upper_estimate": 0.0016993241282350058
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28817.368854593773,
-      "required_pages_per_second_upper_estimate": 0.0022404236742348724
+      "required_token_positions_per_second": 28346.412188495106,
+      "required_pages_per_second_upper_estimate": 0.00220380886496515
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28817.368854593773,
-      "required_pages_per_second_upper_estimate": 0.3696382659867597
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28346.412188495106,
+      "required_pages_per_second_upper_estimate": 0.36359733954791634
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10272979703131568,
-    "producer_p95_seconds": 0.20196048599973437,
-    "gpu_update_seconds": 0.9594213871330809,
-    "producer_batches_per_second": 9.734274075273063,
-    "gpu_batches_per_second": 1.0422948804468233,
-    "producer_headroom": 9.339270736032072,
+    "producer_batch_seconds": 0.08316425607029032,
+    "producer_p95_seconds": 0.2098340830016241,
+    "gpu_update_seconds": 0.9753615313341641,
+    "producer_batches_per_second": 12.024396624852885,
+    "gpu_batches_per_second": 1.0252608575121205,
+    "producer_headroom": 11.728133905385866,
     "sustainable": true,
-    "buffer_seconds": 60.44354738938409,
+    "buffer_seconds": 60.472414942718174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342214.9920353545,
-    "upload_bytes_per_second": 5334357.9514978705,
+    "generation_bytes_per_second": 356178.96931293607,
+    "upload_bytes_per_second": 5301330.690956141,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.43137093348957,
+    "estimated_full_shards_per_hour": 76.42771539250433,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 15955272,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9986490080009389,
+      "bytes": 2653371,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 894000, "loss": 0.6006045937538147, "lr": 3e-05, "grad_norm": 0.31675615674951274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130856451612, "tokens_per_second": 153487.09026421575, "data_wait_s": 1.4489000022877008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894100, "loss": 0.48908618092536926, "lr": 3e-05, "grad_norm": 0.3251328411428243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130870885552, "tokens_per_second": 150828.15868558164, "data_wait_s": 1.60489998961566e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894200, "loss": 0.5504608154296875, "lr": 3e-05, "grad_norm": 0.3310185667273434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130885488264, "tokens_per_second": 151992.03610147812, "data_wait_s": 9.319001037511043e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 894300, "loss": 0.4879351854324341, "lr": 3e-05, "grad_norm": 0.3112756448213933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130900244381, "tokens_per_second": 154333.84903731185, "data_wait_s": 1.7628999557928182e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79a7de021e2e49464f585235281cf2b58a0ea1b758cd9e3c40ac25f3552e6e70
-size 195804
+oid sha256:e90fd808504954e74e71f37ceb06ff6bece2b9e98a566e83cea04a63cf6548c6
+size 197922