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": 75274.92839577794,
-      "required_pages_per_second_upper_estimate": 0.006524640010784232
+      "required_token_positions_per_second": 74636.60581708893,
+      "required_pages_per_second_upper_estimate": 0.0064693118274772535
     },
     {
       "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": 9409.366049472243,
-      "required_pages_per_second_upper_estimate": 0.00375353679969373
+      "required_token_positions_per_second": 9329.575727136116,
+      "required_pages_per_second_upper_estimate": 0.0037217072471422198
     },
     {
       "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": 18818.732098944485,
-      "required_pages_per_second_upper_estimate": 0.004353382472717694
+      "required_token_positions_per_second": 18659.151454272233,
+      "required_pages_per_second_upper_estimate": 0.004316466299096192
     },
     {
       "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": 28228.098148416728,
-      "required_pages_per_second_upper_estimate": 0.001718943075853472
+      "required_token_positions_per_second": 27988.727181408347,
+      "required_pages_per_second_upper_estimate": 0.001704366639845066
     },
     {
       "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": 28228.098148416728,
-      "required_pages_per_second_upper_estimate": 0.002173077135842001
+      "required_token_positions_per_second": 27988.727181408347,
+      "required_pages_per_second_upper_estimate": 0.00215464969618045
     },
     {
       "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": 28228.098148416728,
-      "required_pages_per_second_upper_estimate": 0.3504985056361266
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27988.727181408347,
+      "required_pages_per_second_upper_estimate": 0.3475263193489743
     }
   ],
   "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": 0.08217167489817712,
-    "producer_p95_seconds": 0.2137701259998721,
-    "gpu_update_seconds": 0.9794496198303297,
-    "producer_batches_per_second": 12.169643630109137,
-    "gpu_batches_per_second": 1.0209815591875264,
-    "producer_headroom": 11.919552826980988,
+    "producer_batch_seconds": 0.09743740787484967,
+    "producer_p95_seconds": 0.23670951100211823,
+    "gpu_update_seconds": 0.9878262709411567,
+    "producer_batches_per_second": 10.26299879902817,
+    "gpu_batches_per_second": 1.0123237551145958,
+    "producer_headroom": 10.138059832317566,
     "sustainable": true,
-    "buffer_seconds": 60.725876429480444,
+    "buffer_seconds": 60.25740252741056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289630.39978875464,
-    "upload_bytes_per_second": 5133765.653049201,
+    "generation_bytes_per_second": 319491.1910917171,
+    "upload_bytes_per_second": 5100177.13843864,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.147941544027134,
+    "estimated_full_shards_per_hour": 68.55537223399767,
     "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": 1.4755159030009963,
-      "bytes": 11647558,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 0.8895111490001,
       "bytes": 1197830,
@@ -562,6 +556,12 @@
       "bytes": 806757,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 0.9350021480022406,
+      "bytes": 6160019,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -183,3 +183,4 @@
 {"step": 255400, "loss": 0.5846588611602783, "lr": 0.0003, "grad_norm": 0.13649780259987665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37383657652, "tokens_per_second": 146384.4460172169, "data_wait_s": 1.2704000255325809e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255500, "loss": 0.8631564527750015, "lr": 0.0003, "grad_norm": 0.31558097320959144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37398266478, "tokens_per_second": 149216.92310240545, "data_wait_s": 1.0217001545242965e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255600, "loss": 0.6161860916763544, "lr": 0.0003, "grad_norm": 0.14620557356839445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37412889586, "tokens_per_second": 149742.4806452215, "data_wait_s": 1.537800199002959e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 255700, "loss": 0.6096172984689474, "lr": 0.0003, "grad_norm": 0.17134667809357576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37427631977, "tokens_per_second": 150449.1907204958, "data_wait_s": 1.467999754822813e-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 sha256:5b7b5038047fa9c900772d18b62fd19b38f69c2e8b2103948fd52f6d2aed264e
-size 395127
+oid sha256:4c4de46d2c7c681235f0e105399d676f040a7c02a396422cb440a51cff23e387
+size 397245