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": 74895.5367930377,
-      "required_pages_per_second_upper_estimate": 0.006491755308217934
+      "required_token_positions_per_second": 75169.23250866128,
+      "required_pages_per_second_upper_estimate": 0.006515478559172737
     },
     {
       "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": 9361.942099129712,
-      "required_pages_per_second_upper_estimate": 0.0037346186768508504
+      "required_token_positions_per_second": 9396.15406358266,
+      "required_pages_per_second_upper_estimate": 0.003748266340985583
     },
     {
       "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": 18723.884198259424,
-      "required_pages_per_second_upper_estimate": 0.004331441080160245
+      "required_token_positions_per_second": 18792.30812716532,
+      "required_pages_per_second_upper_estimate": 0.0043472697518926046
     },
     {
       "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": 28085.826297389136,
-      "required_pages_per_second_upper_estimate": 0.001710279466568606
+      "required_token_positions_per_second": 28188.46219074798,
+      "required_pages_per_second_upper_estimate": 0.0017165294539852438
     },
     {
       "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": 28085.826297389136,
-      "required_pages_per_second_upper_estimate": 0.002162124654916207
+      "required_token_positions_per_second": 28188.46219074798,
+      "required_pages_per_second_upper_estimate": 0.002170025850101308
     },
     {
       "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": 28085.826297389136,
-      "required_pages_per_second_upper_estimate": 0.34873196539961926
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28188.46219074798,
+      "required_pages_per_second_upper_estimate": 0.35000635969489774
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08864929903906393,
-    "producer_p95_seconds": 0.14764884700002767,
-    "gpu_update_seconds": 0.9844111299146703,
-    "producer_batches_per_second": 11.280405043691806,
-    "gpu_batches_per_second": 1.0158357312423734,
-    "producer_headroom": 11.104556274955797,
+    "producer_batch_seconds": 0.059018620023435275,
+    "producer_p95_seconds": 0.14108108899995386,
+    "gpu_update_seconds": 0.9808268295343415,
+    "producer_batches_per_second": 16.943805185599345,
+    "gpu_batches_per_second": 1.0195479669686045,
+    "producer_headroom": 16.61893872043894,
     "sustainable": true,
-    "buffer_seconds": 60.049078924794884,
+    "buffer_seconds": 60.81126343112918,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 123,
+    "upload_pages_per_commit": 125,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 682984.713592425,
-    "upload_bytes_per_second": 4844601.51837568,
+    "generation_bytes_per_second": 719313.9375369544,
+    "upload_bytes_per_second": 4611554.76118337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.2763102332571,
+    "estimated_full_shards_per_hour": 77.17401311197983,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 7489968,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.7696208980000847,
+      "bytes": 1062863,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -9,3 +9,4 @@
 {"step": 238000, "loss": 0.5937279835343361, "lr": 0.0003, "grad_norm": 0.1549946949955389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34835903801, "tokens_per_second": 147951.72489235122, "data_wait_s": 1.2591000086104032e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238100, "loss": 0.6234338097274303, "lr": 0.0003, "grad_norm": 0.15404151753059772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34850476923, "tokens_per_second": 147854.77742765663, "data_wait_s": 1.6987999970297096e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238200, "loss": 0.5993404407054186, "lr": 0.0003, "grad_norm": 0.14160709305483027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34865061843, "tokens_per_second": 147445.63180958224, "data_wait_s": 8.603000196671928e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 238300, "loss": 0.6275251917541027, "lr": 0.0003, "grad_norm": 0.1601323362772769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34879729640, "tokens_per_second": 149075.24965014702, "data_wait_s": 1.2643000218304223e-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:f47fc4abecad3b74996aff955b3d0f61e2020581dd5a7bdab006bc28ebadc00c
-size 23463
+oid sha256:fd8ce30ad76fbc896a71c8f77f0b3f0393b1300ab9520d21f8b9099b1c18c2e9
+size 25581