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": 76106.89751131211,
-      "required_pages_per_second_upper_estimate": 0.006596752985112289
+      "required_token_positions_per_second": 76265.51821033117,
+      "required_pages_per_second_upper_estimate": 0.006610501825282774
     },
     {
       "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": 9513.362188914014,
-      "required_pages_per_second_upper_estimate": 0.0037950224145978993
+      "required_token_positions_per_second": 9533.189776291396,
+      "required_pages_per_second_upper_estimate": 0.003802931935651586
     },
     {
       "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": 19026.724377828028,
-      "required_pages_per_second_upper_estimate": 0.004401497825898317
+      "required_token_positions_per_second": 19066.37955258279,
+      "required_pages_per_second_upper_estimate": 0.004410671352670592
     },
     {
       "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": 28540.086566742044,
-      "required_pages_per_second_upper_estimate": 0.0017379415336527574
+      "required_token_positions_per_second": 28599.569328874186,
+      "required_pages_per_second_upper_estimate": 0.0017415637217846709
     },
     {
       "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": 28540.086566742044,
-      "required_pages_per_second_upper_estimate": 0.0021970948679239013
+      "required_token_positions_per_second": 28599.569328874186,
+      "required_pages_per_second_upper_estimate": 0.0022016740156116576
     },
     {
       "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": 28540.086566742044,
-      "required_pages_per_second_upper_estimate": 0.3543723576336596
+      "required_token_positions_per_second": 28599.569328874186,
+      "required_pages_per_second_upper_estimate": 0.35511093446334213
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.29114429542181597,
-    "producer_p95_seconds": 0.3803388999986055,
-    "gpu_update_seconds": 0.9687426818185759,
-    "producer_batches_per_second": 3.4347229731950577,
-    "gpu_batches_per_second": 1.0322658625123713,
-    "producer_headroom": 3.327362744356853,
+    "producer_batch_seconds": 0.11016945710935033,
+    "producer_p95_seconds": 0.36566915400180733,
+    "gpu_update_seconds": 0.9667278441177966,
+    "producer_batches_per_second": 9.076925912482578,
+    "gpu_batches_per_second": 1.0344172934343963,
+    "producer_headroom": 8.774917018591246,
     "sustainable": true,
-    "buffer_seconds": 60.06204627275171,
+    "buffer_seconds": 60.90385417942118,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471060.87863551226,
-    "upload_bytes_per_second": 5244076.132598249,
+    "generation_bytes_per_second": 472137.8518131926,
+    "upload_bytes_per_second": 5209124.225690531,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.07869881915117,
+    "estimated_full_shards_per_hour": 101.3097921924289,
     "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": 1.42470605500057,
-      "bytes": 11031230,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 0.8167821940005524,
       "bytes": 1161839,
@@ -562,6 +556,12 @@
       "bytes": 5902748,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0406464490006329,
+      "bytes": 6219293,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 147500, "loss": 0.6346491239964962, "lr": 0.0003, "grad_norm": 0.14649534405941178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21586278425, "tokens_per_second": 149279.80276100218, "data_wait_s": 1.326200072071515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147600, "loss": 0.6685522776097059, "lr": 0.0003, "grad_norm": 0.15143513756810176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21600793368, "tokens_per_second": 148912.81795938374, "data_wait_s": 1.0970994480885565e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147700, "loss": 0.6182952933013439, "lr": 0.0003, "grad_norm": 0.1372575641211036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21615276738, "tokens_per_second": 148059.53962410265, "data_wait_s": 9.177994797937572e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 147800, "loss": 0.6207619551569223, "lr": 0.0003, "grad_norm": 0.21111374772444552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21629865402, "tokens_per_second": 149524.73886630964, "data_wait_s": 1.1203999747522175e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:496403bc5eb05f345b2bf7c8f81eecd7c70c6a304f5c35fc5d01518097485b1a
-size 697449
+oid sha256:e4a0fe919bd812053436b656bb6becc07d6a1b473373503ab5ac7ec93e0b57e4
+size 699567