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": 74655.20444380539,
-      "required_pages_per_second_upper_estimate": 0.006470923909303247
+      "required_token_positions_per_second": 74246.27787194915,
+      "required_pages_per_second_upper_estimate": 0.00643547919046967
     },
     {
       "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": 9331.900555475673,
-      "required_pages_per_second_upper_estimate": 0.00372263465592615
+      "required_token_positions_per_second": 9280.784733993643,
+      "required_pages_per_second_upper_estimate": 0.003702243790487332
     },
     {
       "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": 18663.801110951346,
-      "required_pages_per_second_upper_estimate": 0.0043175419153377425
+      "required_token_positions_per_second": 18561.569467987287,
+      "required_pages_per_second_upper_estimate": 0.004293892423953473
     },
     {
       "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": 27995.70166642702,
-      "required_pages_per_second_upper_estimate": 0.001704791349390408
+      "required_token_positions_per_second": 27842.354201980932,
+      "required_pages_per_second_upper_estimate": 0.0016954532933576084
     },
     {
       "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": 27995.70166642702,
-      "required_pages_per_second_upper_estimate": 0.002155186611343797
+      "required_token_positions_per_second": 27842.354201980932,
+      "required_pages_per_second_upper_estimate": 0.002143381499037787
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27995.70166642702,
-      "required_pages_per_second_upper_estimate": 0.34761291911080644
+      "required_token_positions_per_second": 27842.354201980932,
+      "required_pages_per_second_upper_estimate": 0.345708856823335
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.243545942335885,
-    "producer_p95_seconds": 0.18490180600019812,
-    "gpu_update_seconds": 0.9875801767510621,
-    "producer_batches_per_second": 4.106001481317458,
-    "gpu_batches_per_second": 1.012576015134079,
-    "producer_headroom": 4.055005668659618,
+    "producer_batch_seconds": 0.09368471524218336,
+    "producer_p95_seconds": 0.1963417759998265,
+    "gpu_update_seconds": 0.9930194767090814,
+    "producer_batches_per_second": 10.674099797548733,
+    "gpu_batches_per_second": 1.0070295935322964,
+    "producer_headroom": 10.599588995302353,
     "sustainable": true,
-    "buffer_seconds": 60.24239078181479,
+    "buffer_seconds": 60.574188079253965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409554.13888470724,
-    "upload_bytes_per_second": 2877880.923329706,
+    "generation_bytes_per_second": 432844.96831240057,
+    "upload_bytes_per_second": 2970521.1586212316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.8807842722503,
+    "estimated_full_shards_per_hour": 92.87845408467305,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 9910795,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 2.0074904029997924,
+      "bytes": 13527574,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -44,3 +44,4 @@
 {"step": 241500, "loss": 0.58491569198668, "lr": 0.0003, "grad_norm": 0.15787964599663515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35348409349, "tokens_per_second": 149601.08783378193, "data_wait_s": 1.3154000043869019e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241600, "loss": 0.6268334165215492, "lr": 0.0003, "grad_norm": 0.14220717144270645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35363108475, "tokens_per_second": 148962.41087267562, "data_wait_s": 9.933000001183245e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 241700, "loss": 0.5771536789834499, "lr": 0.0003, "grad_norm": 0.14329444178759554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35377628896, "tokens_per_second": 148390.99048776756, "data_wait_s": 9.821000276133418e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 241800, "loss": 0.5467068497091532, "lr": 0.0003, "grad_norm": 0.1420090650479915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35392095639, "tokens_per_second": 146137.013868926, "data_wait_s": 1.1516999620653223e-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 sha256:523199c25b35ed5165b26e273c2790cc504a36923995843056bcb485859843ad
-size 98376
+oid sha256:edd325d48385e7037d6f04b4f0bb87e03d0638bc11d5168c2e2985b29b966063
+size 100494